I've been wanting to learn more embedded type projects, and I've been snacking too often so I've been building a box that will only open on the weekends.
I got all the components, tested it on a breadboard, learned to solder and now I'm working on the 3d Print to enclose everything.
I actually just did a test run to see if my current 3d design would fit my PICO board, and it fit, but not that secure yet.
Im a developer but never worked this close to metal, so I've been so happy with how it's been going so far, making me real proud of myself.
I saw your Show HN post a few weeks ago! Really appreciate the smoothness of your UI and the simplicity of your onboarding, I see how much you have dialed in. I've been working on a daily puzzle game too (it's getting there...), maybe you'd enjoy it https://slab17.com/
This is really fun — have you played with making the tile position opinionated (not agnostic)?
i wonder if have the clues point to a starting square (e.g., "E5") would be better than the current "reveal" aid. The spatial information would become more helpful toward the end when the player is dealing with the words they need help on.
There are many language-learning apps, but almost none that focus on improving conversational Hindi for kids.
Made this web app for my nephew, based in Singapore, after watching him struggle to find anyone to practice Hindi with outside of family calls (since most of his friends are Chinese). The idea is to have a 24x7 partner to speak with Hindi and make it fun. This can complement the formal Hindi classes that most kids of Indian diaspora parents take.
For the past 2 months I have been doing a heavy deep dive into image generation and image generation editing capabilities. This then had me discover that you can generate storyboards for short stories, and automate the creation of these as videos with video generation models. This is a topic that interests me heavily, and as such I am now building my own workflows around that. I am documenting the entire journey here:
It's not something I am looking to commercialize, but I actually did drop out of film school (with semesters in creative storytelling) to pursue software 15 years ago. And I feel like this will open up a whole new way of visual storytelling as well as personal and product branding. I have gotten quite some emails about it, from interesting people in different industries, as some more strongly worded (not so nice) emails from someone in the VFX industry since I started. Its by far one of the most interesting tangents I have ever went on.
Currently working on a take on Pokémon GO + Pokémon Snap but for birding. The goal is to explore your neighborhood, find birds, take good photos of them all. Next month, I'll be doing an event to find a rare bird, excited to see how it goes!
I built a chrome extension (with over 600,000 downloads) that lets you chat with page, draft emails and messages, fix grammar, translate, summarize page, etc..
You can use models from OpenAI, Google, and Anthropic.
Built a local-first Kanban board with Tauri (Rust + Svelte) after getting frustrated with SaaS tools and basic offline options. Stores data in JSON files you control, full keyboard-first UX, parent/child tasks, release management, and it's blazingly fast with localStorage + background sync. No telemetry, purely local. Curious what others prioritize in personal task tools. Seems like there's a gap between "todo.txt" simplicity and Jira complexity.
Cool. Love Tauri. So the whole board dataset is stored in localStorage? If you get to a point where the size limitations or synchronous blocking operations are an issue might consider using IndexedDB. There is a nice higher level wrapper around it called Dexie that has full TS typing support and a nice async API. https://dexie.org/
The interface is optimized to let you quickly explore and tweak multiple tints/shades at once so you can customize all colors exactly how you want e.g. try dragging vertically through the saturation curve in one motion to edit all the tints/shades at once, or shift whole curves horizontally by dragging between the dots on a curve.
It uses the HSLuv color space, where (unlike say HSL) the WCAG contrast stays the same when you change the hue and saturation sliders. This makes it much easier to explore accessible colors choices as you know only changes to the lightness slider will impact the contrast. You can also switch from the WCAG2 contrast checker to using APCA, which is meant to correct for inaccuracies in WCAG2, such as it being too forgiving for dark mode color combos.
Note the mobile version is more of a preview and the desktop version has more features.
I probably need to add something like a tutorial as there's a lot going on, but I've added more hints and tooltips recently. Open to feedback on what's initially confusing and what changes might help!
Just launched a startup/life style business where I use AI to help people practice for upcoming interviews - https://hiredcoach.ai
Already have been told by some users that the interview prep they got from it has correctly predicted several of the actual interview questions they got, crediting its prep for their breezing through the interview rounds.
I was tired of inspiration sites like Dribbble full of polished mockups that aren't practical. Or awwward like sites that don't represent the mundanity of most websites.
So, I spent a while building a tool that captures website design snippets. It's now a collection of 4,363 designs from 544 different domains.
For every design, it extracts:
The exact fonts used on the page (so far 561 unique font families I've found)
I've built a self-hosted reddit-like community platform in Go: https://baklab.app
Users can create their own sub-communities, and within them, set up different categories and boards. Posts can be voted on, and board types can include regular posts, Q&A, or live chat. It's like a hybrid of Reddit and Discord but leans more towards a traditional web community. It also supports server-side rendering, making it SEO-friendly. This project is an extension of my previous Hacker News clone, dizkaz (https://news.ycombinator.com/item?id=43885998).
I was getting a little bored of retrocomputing discourse being so centered on gaming, so I'm exploring the productivity software of the 8/16-bit era. I put real effort into learning and using the programs, giving my light-hearted but heartfelt assessment of its form and function for both its time and today.
Using the software inevitably gets me thinking about other things, and I explore those threads as well. For example, "Superbase on the C64" also discusses the legacy and promise of "the paperless office." A couple of other posts got some nice traction here on HN, notably "Deluxe Paint on the Amiga" and "VisiCalc on the Apple 2".
I'm hoping to build a strong monthly readership, so I'm putting in the work. It's been up for two months and five posts now, with a new one coming at the end of this week.
A few of my recent favorites:
- swim lap counter in html/JS that uses the camera to watch you swim and count laps/timing
- video recorder that records your window/desktop and uploads a file to S3
- video conferencing app that allows a 2 year to click on a family member face and initiates a video conference using webRTC, STUN, and browser audio/video capture with automatic bandwidth adjustments (works on all platforms with pure HTML/JS).
- CUDA based ray tracer with HTML UI that can trace over 2m rays per second on my laptop for scientific study, allowing real-time display of optical parts.
- chat front-end for image models like gemini-pro and openai that take other images and text as references and generate a big library of options to chose from in seconds, I've been using photoshop for decades but I tend to use this more now.
I'm curious if you mean they're running a raytracer on the back end, and you interact with an HTML UI, or if it runs browserside, maybe via WASM. AFAIK CUDA isn't directly compilable to WASM (yet?)
I have a node middleman that proxies request from an HTML/JS front end to a native cuda process using web sockets. To support multiple windows, the node process process provides communication between two browser windows. This lets me have render a model using 3JS in one window and a ray traced version in another window.
Discovered in-door bouldering / rock climbing and now go 3x a week, am absolutely loving it! Because of that, I haven't really worked on any side projects in a while. Perhaps I don't need to? My job advances me plenty in my field, but it is a bit of a bitter-sweet feeling in a sense, like maybe I should try to squeeze more out of my free time somehow.
I climb a lot! (Actually currently sitting on Big Sur ledge on el cap posting this).
It cuts into my free time programming for sure, but imo super worth it! Enjoy it, it’s a wonderful hobby.
I’m replying from the cold east coast (from the edge of a wood chair in a lovely iykyk type of restaurant) to a human posting from el cap on hn; We have achieved peak technology. Oh yeah, I’m working on urban logistics, powered by AI.
I’ve been hesitant for fear of injury harming the ability to type, but might give it a go in the spring. Thanks for mentioning this I’m inspired to try it finally.
Couple things to avoid finger injuries: go easy on one- and two-finger pockets, use an open crimp whenever possible (all finger joints are bent the normal direction, and your palm/thumb aren't really involved), and don't bother with the hangboard or campus board for the first ~year.
I wouldn't worry about it too much though - almost all of the people I know with finger injuries were trying to push into really being competitive climbers, not just doing it casually for fun/fitness.
Oh also to keep from tearing your skin don't climb tired. (That won't keep you from typing, it's just painful.)
I'd like to add to this that do not make any food with chilli peppers like habanero or such if you just came from the gym with torn skin. I found out the hard way.
I struggled with hand and wrist pain for years from spending too much time at a computer. I did physiotherapy for years and while it helped me manage pain, I was never able to truly build enough strength to get ahead of it until I started bouldering. I took it very slowly—I spent months on very easy problems—but because it was so much fun, I kept going back. Initially, I would only go on Saturday mornings, so I had the full weekend to recover before jumping back into the work week on Monday. After a two or three months of that, I was able to climb anytime I wished. I'm still not a particularly advanced climber, and I typically only go once per week, but I am still slowly progressing, and I absolutely love it.
I’ve been climbing for 20 years and it’s the thing that prevents RSI for me and makes it possible to use a computer too much :). Certainly possible to injure fingers but would be a very rare climbing injury that would threaten coding.
Climbing easy routes in a gym is pretty low impact. It’s only when you start to move into really hard crimps or slopers where you’ll hurt yourself. I was a climber bum for years and have climbed crazy stuff around the world and never hurt myself to where I couldn’t type. A lot of bloody tape, but still able to type.
Try top rope climbing! Bouldering is injury prone because every fall is a ground fall. With top rope climbing you should never hit the ground so way less injury prone.
I've been working on a 3D voxel-based game engine for like 10 years in my spare time. At this point it's getting pretty close to being shadertoy for voxels.
I also wrote a metaprogramming language which generates a lot of the editor UI for the engine. It's a bespoke C parser that supports a small subset of C++, which is exposed to the user through a 'scripting-like' language you embed directly in your source files. I wrote it as a replacement for C++ templates and in my completely unbiased opinion it is WAY better.
Working on Spine AI, a visual workspace to think across multiple AI models.
You can chat, branch, and connect 300+ models on an infinite canvas: useful when you need to explore tradeoffs, check blind spots, or generate assets (research, slides, prompts, images) from the same board.
I have been working on Buckaroo - my table display library for dataframes in notebook environments. Buckaroo adds table and analytics features like histograms, summary stats, sorting, and search to every dataframe. Recently I have been working to make it work better with large datasets.
This involves making it lazy for polars, allowing it to read arbitrarily large files no longer requiring loading the entire dataframe into memory. When a large dataframe initially displays, no summary stats will be available. Summary stats are computed in the background in groups of columns. Then results are cached per column. To accomplish this I wrote a polars plugin in rust that computes hashes of columns. Dealing with large data like this is tricky, operations sometimes crash, sometimes take all available memory, and sometimes they just run for a very long time. I have also been building an execution framework for Buckaroo. It uses multiprocessing based timeouts, and the caching to execute summary stats in the background.
Being able to control the execution, recover from timeouts, crashes and memory exhaustion opens up some interesting debugging tools. I have written methods that take arbitrary groups of polars expressions and produce a minimal reproduction test case through a git-bisect like process.
All of this assures that if individual columns of a dataframe fits into memory, summary stats will be computed for the entire dataframe in the background. And because it is cached, the next time you open the same dataframe, the stats will be display instantly. When exploring data I do this in an adhoc way manually (splitting up a dataframe by columns and rows), but it is error prone. This should all be automatic.
I will be presenting this at PyData Boston in December.
The Column's the limit: interactive exploration of larger than memory data sets in a notebook with Polars and Buckaroo
Today, I am implementing tag-triggered webhooks so they can get triggered based on the tags assigned to messages.
Use case: for example, I want to auto-tweet all my advice messages and auto-create linear tasks for `company` todos, auto-start a cursor agent when I tag messages with `Cursor task`
For some reason, I got 16 users out of nowhere, so I added a landing page last week. The connector marketplace is next, so that I can share these sub-connectors with the other users. Need a couple more weekends.
Over the years, I've read countless books. I started documenting one idea that shaped my thinking from each of these books. This idea may or may not be the core theme of the book.
A Civil 3D plugin (Genabler) that will include all the network catalogs and collate the Civil 3D styles for civil engineers to use.
There are some out-of-the-box catalogs and styles shipped with the default installation, but they are quite limited and fairly well hidden—which is not surprising, given that Civil 3D is a huge beast. As a result, they are not commonly used.
When people think about Civil 3D, they often assume it requires BIM modelers (in a sense, just glorified drafters) to create all the necessary catalogs and styles, and to assist with their use.
My Civil 3D plugin will:
1. Make standard, market-compliant catalogs and polished styles available to engineers at large. Think of it as the WordPress theme provider equivalent.
2. Make the entire process easy and painless through the plugin, with prominent buttons for quick access.
If the plugin is done well, there will be less need for BIM modelers, since for a fee, engineers could simply purchase catalogs and styles that are so easy to use they require no technical training.
As a side benefit, I also get to explore how LLMs can help me write code. It has been a while since I last updated my AI usage policy [0], and I look forward to revisiting it.
Working on https://outcrop.app, a knowledge base for software teams with instant search, realtime collaboration, and LLM-driven workflows. It's built using Rust and friends. I'm looking for more early testers! :)
It's an explorable database of films, TV shows, books and board games based around their historical setting: where and when the thing is set. It's been incredibly complex and interesting getting the (messy) data, making sense of it and trying to design a UI to explore it.
Yeah, that's a whole different thing. I saw this map once that showed places in California and how they were used to film various locations around the world. Turns out LA is very well situated.
I’ve been working on https://canine.sh which is a free, open source Heroku alternative for 2 years now.
It’s exactly the product I wish I had when I started my previous company. Running on PaaS is incredible for devex but the pricing is bonkers, and the vendor lock in makes it really hard to deal with annual price increases. We spent close to 400k / year for just 128GB combined fleet in our last startup on Heroku.
Canine tries to get the best of both worlds: developer friendly PaaS with no lockin or price gouging.
Just added build packs as a build option recently.
Also got a sponsorship from the portainer folks which lets me work on this close to full time
Hoping this saves someone the headache I had two years ago.
All of the street and satellite tiles are thanks to maps.black. The search uses Nominatim's sqlite3 mode. I was told that it's experimental only because it hasn't been tried in production yet, so I'm sort of testing it in the process. So far I'm only doing administrative boundaries and natural features, but so far so good! I'm going to slowly add a few more types of POIs, I just don't want the database file to get too big.
Note that Internet in a Box has an OSM offering already, but the data is five years old and the tech makes it harder to update. As of today, there are much easier options on the table, and we get cool stuff like 3d buildings. Also, the search was much more limited.
I'm building a coding agent, named VT Code [0]. VT Code is a Rust-based terminal coding agent with semantic code intelligence via Tree-sitter. Supports multiple LLM providers with automatic failover and efficient context management. Support OpenAI, Anthropic, Google Gemini, xAI, DeepSeek, OpenRouter, Z.AI, Moonshot AI, MiniMax, and Ollama (local & Cloud). Agent Client Protocol and Model Context Protocol fully support. VT Code supports a rich set of configuration options, with preferences stored in vtcode.toml. Has both Visual Studio Code and Open VSX extensions so that you can install in VS Code or Cursor, Windsurf, Eclipse.
I've been building it for several months now and enjoy the learning process, I also wrote a blog post and learnt a ton about terminal, ANSI processing. The learning has been immense for me, I now have working knowledge of ANSI escape codes, grapheme clusters, terminal emulators, Unicode normalization, VT protocols, PTY sessions, and filesystem operations, all the low-level details I would have never think about until I were implementing them. [1]
It looks inside each file to see what it’s about, then moves it to the right folder with a single click.
Everything happens on your Mac, so nothing leaves your computer. No clouds, no servers.
It already works with images, Office (Word, Excel, PowerPoint) PDFs, ePubs, text, Markdown, and many other file types (30+) in English. Next I’m adding multi-language support.
If you have messy folders anywhere on your Mac, Floxtop can help.
It's very interesting because the Ruby codebase uses a `typedef uintptr_t VALUE` type to mean any of the following:
- A pointer to the heap
- A Ruby tagged value (which may be a pointer to the heap)
- Any integer value that fits in `uintptr_t`
Fil-C doesn't allow you to carry around pointers using integers, in the sense that when you do that, the pointers lose their capabilities.
But in Ruby's case, it's not as simple as changing the typedef to a pointer type, since `VALUE` variables often end up being used for integer math (including bit math, shifts, etc).
So, it's going to take a nontrivial patch to Ruby to get it to work in Fil-C. I think I'm about 70% of the way through (I started Friday afternoon).
I recreated a little tool to simultaneously mount all the commits in a git repository as directories at the same time (but re-use the same inodes for the same content).
The original was in Python and actually had a decent excuse for existing for a very specific problem at work a few years ago. The new version is in Rust and exists just for fun.
This was also a small experiment in coding with OpenAI's codex. I wrote the Python original by hand---like a caveman. Codex was mostly ok at the actual code, especially once I told it to make `cargo clippy` happy, but it needed lots of help with the design. It kept insisting on extra complications and state.
But perhaps I'm a bit unfair here, because I only figured out the nice and simple design after reflecting on the connection between Linux's fuse and git's design for a while when writing the original. So it's only fair that the computer would also need some help to see how to match them up nicely.
I’ve always wanted a typing application that’s both more than typing random words and is data-focused so I built this.
The more you type, the more the analytics system learns about your typing patterns and generates natural text to target those weakpoints (SmartPractice mode).
There’s a lot of variety as well; you can practice typing code in any programming language, or type text of various topics, use custom text, etc).
It has been a super fun experience so far - I'm using CPLDs instead of an FPGA which makes the logic a bit more era period. I have a working system now with the math coprocessor, SRAM, DRAM, and other device support.
I am just about ready to get the VGA card I designed produced so I can work on debugging the design.
While this is fundamentally a system that ss less powerful than my apple watch, it is just fun to work on. Going back to very first principles debugging, building tools, and of course getting to exercise an old logic analyzer!
I'm making Easel, a 2D game programming language designed to match how humans, not computers, think about game logic. It also has automatic multiplayer. I've been working on it for 3 years!
Easel feels like a declarative programming language even though it is imperative, because lots of useful game-oriented features are first class. Like behaviours - you just say `on Pointer { ... }` and you have a concurrently-executing coroutine that's lifetime is managed. But you don't think about any of that complexity, you just think of your entity as having a behaviour and go forth and make your game.
It also happens to have automatic multiplayer. Normally with multiplayer you have to worry about doing everything in a "multiplayer safe" way (i.e. be deterministic and only modify the things your side has authority over). My idea was to put all the multiplayer stuff in the programming language itself, underneath all your lines of code. This way, anything you write in that programming language can just be made multiplayer, automatically. So you can just pretend all your players are in one shared world, like a singleplayer game, and the engine does all the multiplayer for you. It was really difficult to make but it makes multiplayer so easy for you now.
Easel is my idea of how games should be made, or at least as close to the idea as I can achieve with 3 years of work, and I would love for more people to try it out.
I would love to hear more about what you were trying to do with your project before. Was it more similar to the declarative coding part, the automatic multiplayer part, or something else? Part of why I'm doing this is to explore the design space of how games should be made and I'm interested to hear what problems, issues, pet peeves, "bugbears" etc that other people think are worth solving.
It's been a while. But I believe what caused me the most headache while trying to build something like this was handling the interactions between different elements. Declaring which objects were affected by "attacks" or could be "player interactive" or "affected by player but not by NPC". Really this boiled down to proper inheritance. But I found myself so deep and tangled a fresh reset would have been better. Then determining if the object itself or an "objective manager" should perform the calculation each cycle.. etc
It was messy. I ended up having NPC, Item, Attack classes and for each a NPC Manager, Item Manager, and Attack Manager to calculate all their interactions and states.
That's why your project seems interesting because it seems to handle the heavy lifting of behaviors and "behind the scenes".
I’m working on https://regularly.co/ - A website made for inquisitive minds to get their daily puzzle fix. Still very much a WIP (mainly working on tuning the difficulty of puzzles to make it enjoyable for most). That being said I really do enjoy the unique combination of puzzles when I do them each day. I’m looking for feedback so if you do take a look please do let me know your thoughts!
Incredible. Thank you for sharing this, I love puzzles and like setting aside some time in the morning to do them. This will enhance that habit so much! A whole slew of daily puzzles, I'll let you know how it goes!
At the intermediate level lots of learners struggle to find suitable content that matches their level and interests, more than a few learners turn to notebookLM podcasts to provide that, but that's a bit of a hassle to set up. So I built a platform that generates and manages infinite and shareable streams around your interests or specific vocabulary. It also provides live interactive transcripts (karaoke / teleprompter style) if you need it.
Core features work but still rough around the edges. Happy to help you out with any issues you encounter, languages to add, feature requests etc...
I suspect AI company want improved efficiencies and developing a framework that can be applied in determining the minimal-energy, maximal-efficiency architecture for ai models. Calculating the precise limits, like a Cognitive Event Horizon, where a model becomes so complicated it literally costs more energy to run than the knowledge it provides, and the Semantic Horizon, where it simply gets too complex to be accurate, etc. Lots of cool implications such as around a fundamental mathematical maximum learning rate which results in trying to get anywhere close to that that by doing stuff like aggressively filtering of the data.
I'm building a small rural ISP and web hosting service, as a way to learn about low-level networking stuff. I've got an ASN + IP space, and am working out the details with a colo, local fiber provider, and some upstreams. Right now I'm configuring the hardware itself (server, router, switch, etc) and learning all the bits and bobs (Proxmox, BGP, OPNsense, IXPs, etc)
All I can say is good luck. We spun up a co-op isp to take advantage of fiber grants for rural areas about a decade ago.
Maybe it was because of the grants, but it was a fucking nightmare getting off the ground even though we had nearly 90% of the population in three counties on board for the co-op. The red tape and regulations (in our state at least) made it clear that government runs for urban and suburban interests and actively undermines rural needs. I'm talking government in bed with large providers who had exclusive rights to run "high-speed" Internet to our towns and farms, even though they had never and were never planning on anything above dsl for most people and cable for the ones in town.
If I was more charismatic (and wasn't 1000% sure there were pictures of me doing drugs when I was in college), I would consider a run for state office, because it's a shit show for small towns here.
And that's the story of a) when we got sued by a large provider that I hope goes out of business and burns to the ground, and b) the last time I volunteered on a large project and why I will never take the lead on anything bigger than the Lion's Club pancake breakfast now.
I'm funding this myself, and my current approach (hopefully!) avoids most of the red tape. I'm leasing fiber from a local ISP for the colo <-> my home connection, and once I have myself as a successful "customer" of my own ISP, I'll start doing the last mile build out, which is where I expect the red tape to begin.
But I haven't decided if I'll do fiber or wireless, and if I go wireless, I might be able to avoid pole agreements entirely by just working directly with my neighbors. The problem is that our area is pretty heavily wooded, so I'm not sure if I can place antennas high enough to cover a reasonable swatch of the area.
> If I was more charismatic (and wasn't 1000% sure there were pictures of me doing drugs when I was in college), I would consider a run for state office, because it's a shit show for small towns here.
Loads of politicians have come back from worse! Don't let that hold you back.
I'm making an app for self-tracking. Combining elements from habit trackers, health logging and journaling. Built for rich customization and local-first. Want to be free of rigid structures of many existing apps while providing a better UX / usability than using a spreadhsheet.
I'm trying to learn 3D scanning and printing: I have a few small projects that I want to do to develop the skill:
I want to 3D print a shell that goes over my car fob: I keep leaning on it and setting off the alarm. The shell would make sure the buttons never get pushed.
I want to 3D print a sleeve that keeps the NCAS dongle in my car charger. I really wish there was a dongle that stayed attached with screws or similar.
I’m working a Garmin watch app to query all the rich data on the watch (health, physical, environmental, location sensors) from the watch + general AI assistant. Privacy focused using your own keys and Gemini. API calls direct from watch - no backend. https://untether.watch
Building vet. The goal is to automate open source package vetting beyond just CVE but actually identify code capabilities, malicious code and other security sensitive attributes through code analysis.
I'm working on Flavia, an ultra-low latency voice AI data analyst that can join your meetings. You can throw in data(csv's, postgres db's, bigquery, posthog analytics for now) and you just talk and ask questions. Using cerebras(2000 tokens per second) and very low latency sandboxes on the fly, you can get back charts/tables/analysis in under 1 second. (excluding time of the actual SQL query if you are doing bigquery).
She can also join your google meet or teams meetings, share her screen and then everyone in the meeting can ask questions and see live results. Currently being used by product managers and executives for mainly analytics and data science use cases.
We plan to open-source it soon if there is demand. Very fast voice+actions is the future imo
What kind of plan do you have with Cerebras? It seems like something like that would need one of the $1500/month plans at least if there were more than a handful of customers.
Great feedback thanks! We have added a synthetic e-commerce dataset as an example when you sign up so you can test it without your data first. Will also add a demo video ASAP.
I've been writing https://urbanismnow.com weekly for a year. The idea is to bring you the best ideas from around the world to inspire action where you (c)are.
It's been going well for a side project and now I'm thinking of expanding to have a directory of urbanists on a map so you can easily find people involved in the local discourse and how to get involved.
I recently have gotten into the "drag and drop" forms of programming like Node-RED and n8n.
Obviously, anyone here who has read my posts knows I know how to write code, but having a bunch of built in connectors that are agnostic to each other with the Oauth and the like being somewhat plug and play allows me to iterate on some ideas a lot quicker.
I installed an n8n instance on my server, and have become kind of addicted to making different Discord bots, and I'm having more fun with this than I thought I would. 95% of the stuff on there is basically drag and drop, and when I need more elaborate logic then I can easily drop into JavaScript. I am looking into writing new nodes for different services, and I keep having new ideas for different stuff I want to build.
A place to find great blog articles by regular folks related to dev/tech world.
Wondering about the best way I can add a weekly newsletter built on top of the content currently being ingested and still looking for more sources to add to the database (let me know if you have any good recommendations).
I was a YC founder in 2006 and now work as a data scientist full-time, but on the side I also do Christian apologetics, helping fellow engineers/scientists/mathematicians seek answers to life's deepest questions.
OpenRun runs as a web server, which does GitOps driven app deployments. You can currently deploy apps on a standalone machine, on top of Docker/Podman. Working on adding support for deploying on top of Kubernetes. On Kubernetes, OpenRun will replace your build jobs (Jenkins/Actions etc), CD (ArgoCD etc) and IDP (Backstage etc). The same declarative config which works on a standalone machine will work on Kubernetes, with no YAML to maintain.
Today I'm hacking on automate-terminal, a command line program and Python library that abstracts the various terminal emulator automations (iTerm2, WezTerm, Kitty, tmux) into a single API. Mostly made for use by other tools. https://github.com/irskep/automate-terminal
I'm working on fighting IBM's patent trolls.
IBM slapped the words 'AI Interpretability' on Gauss' 200 year old continued fractions and was awarded a patent.
Now they can charge rent if they encounter a continued fraction library in the wild.
The authors simply implement a continued fraction library in Pytorch and call the backward() function on the resulting computation graph.
That is, they chain linear neural network layers and use the reciprocal (not RELU ) as the primary non-linearity.
The authors reinvent the wheel countless times:
1. They rename continued fractions and call them ‘ladders’.
2. They label basic division ‘The 1/z nonlinearity’.
3. Ultimately, they take the well-defined concept of Generalized Continued Fractions and call them CoFrNets and got a patent.
IBM's lawyers can strip out all the buzzword garbage if they feel litigious and sue anyone whose written a continued fraction library.
Because, that's what the patent (without all the buzzwords) protects.
You sent me down a rabbit hole. In trying to track it down for myself I read a couple of others that I thought might be it, and was stunned by how obtuse these patents are.
What sort of leverage does this stuff provide? You mentioned "charge rent". What does that look like?
Honestly, I don't even know where to begin. It's insane IBM owns the patent to continued fractions.
If you wrote a continued fraction class in Pytorch and called backwards (or even differentiated the power series) then you're infringing on their copyright.
Every time I talk acquaintances, friends and family members about finances I'm always shocked at how little people know about basic things like tax brackets, 401Ks, IRAs, ETFs, compounding interest, debt management and etc. So I decided to write a financial literacy/education book with a bit of humor and easily comprehensible language to distill some of these topics. I'm about 1 month into it and try to write a chapter a week.
I'm working on https://mimicmarketer.com It allows you to define different personas that you can then test marketing on. This allows you to see how different personas will interact with your marketing. Currently, it has a feature that allows you to define basic personas and test them against two types of copy, as well as a tool that grades your email subject lines and bodies against a generic persona, assessing the likelihood of user interaction with the content.
My other project is https://eggexplorer.com This is a site I wish I had when building out my flock of chickens. It allows you to see the different characteristics of chickens and which hatcheries sell each different breed. You can also see which hatcheries sell hatching eggs for each breed as well.
I really liked the concept of games like cards against humanity, quiplash, whose line is it anyway etc. However, there was no virtual way to play it with a group of friends. Quiplash required steam setup (which was not possible on my corporate mac). So i built this as an alternate to build upon the formula.
After creating the feature request for Claude Code hooks[1] a few months back, Cupcake is nearly ready for release.
Cupcake is a governance/policy-enforcement layer for agents. Its innovation is binding OPA/rego to agent runtimes (via hooks).
I do not believe we will every strictly rely on "better" models in the wild without deterministic guarantees or ways for enterprises to factor in their own alignment - system prompts dont cut it.
I'm resurrecting peer-to-peer Matrix (https://arewep2pyet.com) thanks to the Dutch government, who started funding it in October.
The main question is which P2P overlay network to use, if any: the prior incarnation used Pinecone (a variant of Yggdrasil), whereas this time we're pondering keeping it simpler and more scalable and using Matrix itself as the backbone to connect together smallish local P2P meshes - so by default you try to route via Matrix, but failing that you look on your LAN or BLE to see if you can talk directly to whoever you're addressing. Time will tell if this works :)
We are open to suggestions :) And the very first generation of P2P Matrix was indeed built on libp2p (and Protocol Labs led Element's Series B). However, the thought experiment here is whether we can get away without a full global P2P overlay at all in the interests of keeping it simple & stupid. We might well end up back at libp2p tho!
- Local-first app for comparing hardware builds, down to the individual component feature level: specs, benchmarks, even cpu extension support, lanes, how many speakers in X laptop, dolby atmos? screen panel manufacturer(s), etc. Basically, no-nonsense real product comparison for transparent and fast decisions.
My friends and I have been hacking on http://dateit.com for a while.
It's an event planning app (works best on iOS and android, but there is a web app) with lots of fun features:
We started working on this all the way back during the Covid lockdown when we wanted to capture that "facebook events" experience without the facebook.
It's grown into something much more than our original idea. Most of the features are free and we have a fair pricing model that doesn't nickel-and-dime you like many of the competing apps do. Would love your feedback!
I am finally making my own blog. I have been only planning for ages. I found that I had a lot to say for the past years working on AI, and I want to record them somewhere. I do not expect a lot of visitors or at all in fact. The blog is going to be just for me to remember stuffs and to keep track of them.
I am using hugo to build suckless static pages. LLM helped me so that I don't need to read all their docs. I haven't finished it yet nor posted a single blog. But there will be one soon.
Currently working on getting back into a fitness routine. I got into this habit of hacking on side projects in my very little spare time but I have realized taking care of my body will pay off far more than any project
All the best - the best project to work on!
I recently tested a walking pad with a standing desk at my friends place - I was surprisingly productive being able to walk while using the computer. Will be investing in a standing desk and walking pad for my home office.
I'm working on a boardgame with the help of AI. It's way too easy to create placeholder art with an n8n pipeline, but GPT-5 regularly fails at writing and debugging LaTeX which I'm using for all of the card creation.
Specifically, TikZ is often outside the ability of GPT5 to successfully write or debug.
I'm working on a book about using WebViews for cross-platform music software GUIs. It has a particular focus on performance, which I gave a talk about at the Audio Developer Conference last year:
Experimenting with various AI models via GitHub Co-Pilot on an extremely niche project to see how far these models have progressed. Used like ~60% of the premium quota to develop the following projects:
Manual audio splitting tool for the above project: audio-splitter-6b3.pages.dev/
I've always been skeptical of AI-generated code. This is my first experiment with AI agents, where the full code base, implementation, debugging, and deployment are done using AI Agents MCPs.
Used VS Code all the way, i.e., all the source codes, including the code to generate the Google Play Store APK. I only reviewed the source code before committing and helped debug by suggesting ideas/algorithms.
Mostly used Claude Haiku 4.5 like 75% of the time, where it failed, switched to the sonnet 4.5 or GPT 5 codex. Interestingly, when debugging, sometimes one model struggled even after numerous iterations/feedback loops, but then the problem was solved instantly as soon as I switched to another model.
Initially, I thought the audio splitting could be done automatically using some AI models from Hugging Face or Whisper. But the audio files have some complex repetitions; the output was miserable.
So, for now, this splitting is done manually using a Web UI (The audio splitting tool splits the large audio files into multiple small audio files, think of it like a long paragraph is split into multiple sentences.)
I will attempt again to automate this splitting task using AI, after drafting a game plan for tackling the challenges. I'm thinking of using energy drops and other similar factors to create segments.
If you play DnD, I would love feedback! Feel free to leave it as GitHub issues or discussion.
If you don't play DnD, you might still find the repo interesting. It's hono on bun, I render jsx server side and client side is all htmx. I use vercel's ai toolkit for the LLM interactions, which are super fun and work really well. I think this is a great use for AI actually. I've structured the code so the same services can be called either by the user via forms and routes, or via LLM tool use, so for every action in the code you can do it via either LLM or "manually".
The LLM usage is fun and interesting. What model are you using, and how much customization are you doing to integrate with the app and maintain character?
I suggest adding an export function to make the characters more portable. Maybe export to PDF as well as JSON.
Working with a group of friends on a "microcontroller-for-makers" kind of thing called the MakerPort. (https://makerport.fun) Sort of similar to an Arduino or micro:bit, but uses the MicroBlocks programming editor (https://microblocks.fun) created by John Maloney, who was the original team leader for Scratch at MIT for 11 years. The hardware includes an mp3 player, I2C ports, accelerometer and true capacitive touch sensors.
I continue to work on My Financé, my personal finance tool.
I’ve been struggling to find substantive traction, so I’m trying to niche down to make the tool really helpful for people who want to quit their jobs.
I built a rudimentary planning and forecasting engine, and am trying to run paid ads to see if the signals resonate with people. I don’t love ads, but maybe trying to understand them will further inform my opinion on them.
One thing I would love to come up with is a way to make the app fully local first, while continuing the ability to sync accounts via plaid. It would be great to not be able to see people’s data at all. Im trying to figure out if there is a good user experience I could provide while minimizing the amount of data I actually have access too. Maybe this feature won’t matter to my primary customers though, I’m not really sure.
I still have a ton of fun working on it, and if it never really makes any money I consider it a great success for my personal learning.
Building a docs website [1] for my speech-to-text CLI tool, hns. I use it 5-10 times daily to transcribe my voice, and a few developer friends I've shared it with have also adopted it for daily use. They like that it runs in the terminal and keeps all data local. So, felt like I should write down guides for new users to get started quickly and to highlight key use cases.
Building this documentation website using Docusaurus. This is my first time using Docusaurus, and it feels like a very nice tool for quickly developing a documentation website.
World model for AI agents. Doing process mining of missions, operations and logistics to transform them into digital twins. AI agents can then leverage these digital twins as world models for control or prediction.
I kinda gave up on building apps for Rocknix (since there is no easy way to distribute software) and instead have been looking at my apple watch. I ported over some software for workouts that is streamlined. I'm working on a way with MDNS to sync data to my Linux PC automatically when I'm at home.
If it works out, maybe this could be a way for me to replace the compromised Apple Music app with something that actually syncs to my music on my desktop.
FURS does for Forth, what headers do for C, namely provide all the embedded configuration information inside a Cortex-M MCU, for the up to 100 inbuilt peripherals.
Without this data, neither C nor Forth (or any other language) have any clue about how to use the peripherals.
FURS does this by intercepting the Forth user source as it's uploaded to the on-chip compiler and transforming it into language the MCU inherently understands.
The Forth user source code is not altered in any way.
I've used the Fossil DCVS for the entire FURS project so that all the flowcharts, pictures, code, user doc, trouble-ticket, wiki ... everything is contained in the ONE FILE, under 5MB.
This one file gives you a web server so all you need is a browser to easily view all the above from the main menu.
I am working on PocketWise (https://pocketwise.app) a lightweight personal finance tracking app. Goal is to make double entry accounting simple and approachable for everyday use. It’s my first project of this kind, so I’d really appreciate any feedback.
I’m building https://unrav.io : A tool to fight information overload.
It lets you turn any article, YouTube video, or PDF into summaries, mindmaps, podcasts, chat conversations or infographics that match how you learn with just one click.
We just launched this week the Chrome extension so you can do all this in one click on any page, no login needed (with generous freemium usage).
I'm making a game finally! Merge-three + village sim.
Hoping to actually take this one to something polished as opposed to the many half-finished prototypes littering my git repo over the years. I've discovered (always knew?) that heavily cutting scope is the best way, and been successful thus far.
It gets pretty boring/unbalanced by ~150 turns, but I have some ideas on how to fix. I'm still playing with ways to help de-clutter the board and make use of the economy aspects.
After some false starts with ai-gen art, I had fun learning to color the pixels myself. The process wasn't as scary as I'd thought and the results are better than I hoped.
A kernel extension-less sshfs for macOS. I tried using FSKit and got halfway before I felt too constrained by the extension security model (must be app sandboxed, must be approved by the user in system settings). Now it’s just a standalone command line binary that doesn’t require any special permissions since it proxies NFS to SFTP. Everything “just works” and performance is reasonable
Check out my project at https://www.MobiusClock.com: A 3D WebGL Clock on a Möbius Strip that shows 24hr time on a 12hr face. The hour indicator follows the edge of the strip, thus must make 2 turns to return to its starting point, giving you a 24 hour clock. The minute and second indicators move along the middle of the strip and thus return to their starting points in only one turn. Has the ability to rotate!
Just added a new feature: a 'Fast Mode' button to temporarily speed up the hands, which helps visualize how the slow-moving parts work, how the hour indicator moves along the edge. Would love feedback on the implementation.
I'm thinking a lot about the ARC-AGI ML benchmarks, especially the "shape" of the dataset and what that says about how it should be solved. I think there's good reasons to believe that deep learning - at least differentiable SGD backprop style - is a bad fit for this specific benchmark, due to the tasks being almost entirely discrete symmetries, and also having so little data to approximate the discrete symmetries with continuous ones (considering deep learning to be the learning of continuous symmetries). I think that a more explicit and discrete approach is the way to go, and it's possible to build something surprisingly general and not heuristic-based even without gradient descent, guided by minimum description length to search for both grid representations and solver functions. I'm looking for teammates for ARC-3 so hit me up if this sounds interesting, I'd love to chat!
I made a viewer on my website to build intuition for my preferred perception algorithm which is entropy filtering + correlation. Pretty neat to check out the heatmaps for random tasks, there is a lot of information inherent in the heatmap about the structure of the task: https://synapsomorphy.com/arc/
I've been working on a sillier project lately. Green teeth!
Lumina has made a probiotic strain that is able to, theoretically, prevent cavities. I don't care that much about, but I do think it is a neat strain that can likely colonize your mouth. I'm genetically engineering it to express sfGFP, which would theoretically make my teeth fluorescent green under black light. Would be fun at raves! Also, if I make out with anyone, you could theoretically see changes in microbiome composition just from green-ness. I do wonder how much microbiomes are shared while kissing: this would be an example of a way to directly measure that, instead of just measuring on proxy like much microbiome research
I'm working on boiling the ocean - we're building a new CRM to compete with some of the big players. I tried very hard to avoid doing this, but I've helped enough business owner friends set up CRMs to realize there's MUCH to be desired. My goal is to create a CRM that people rave about - something that is very rare. Pretty much everyone I help views CRMs as a necessary evil. Our bold challenge is - can we make a CRM that is delightful to use?
Of course we have to slap "AI" on it in this market, but we plan on adding AI features that are actually thoughtful and not just a glorified chatbot.
i'm currently unemployed but i have experience as a CSM and account manager for multiple companies (all have used Salesforce) if you ever want feedback or bounce ideas.
This would be awesome, we're in hardcore MVP build mode right now - when we get closer to launch I'll hit you up.
I'm a little delusional, but I think there's ground to steal back from Salesforce. Most folks I talk to hate how complicated Salesforce is (one even calls it Salesfarce). I've heard a story or two about smaller companies trying to adopt it and wasting a hundred thousand or two implementing Salesforce only to have it never get used. On top of that, you need to train your employees to use Salesforce effectively.
The key is simplicity, building a CRM that anyone can instantly understand just by looking at it. This is insanely hard but I think we'll pull it off. I'll show you more what I mean when I reach out. Thanks!
Sounds interesting! As someone who works in Revenue Operations the CRM space is ripe for disruption, especially around using CRM data to help sales teams explore the data.
I vibecoded a POC of what something I think would work (around the latter part around exploring the data). Need to complete it and start testing it.
Thanks! I totally agree. It starts with getting the basics right. I think this is where most CRMs get it wrong - getting users to correctly enter and maintain clean data is a challenge. We're trying to build smarter schemas for contacts, quotes, sales, etc. but we're also making a really intuitive UI for easily updating information. Further still, we're trying to automatically fill in data for the customer. Many fields can be automatically inferred based on the context of a deal.
Once we get this right, I think the next step is exactly what you said, building really good tools to explore the data. Making it easy for non-technical users to run machine learning on their data to make business decisions or see cool visualizations. We realize that so many of our customers want to know this stuff and have no way of getting at it! We've got a lot of ideas for both visualizing and analyzing the data. I think there's a ton of potential for cool things here. Heatmaps, spiderwebs, interactive charts, etc. Stuff that brings the data to life. One of the common asks from some of our early customers is a heatmap of the world to visualize their sales/reach and see changes over time. Visualizing progress per sales region, etc. I think sales especially has a lot of opportunities for better lead generation and qualification as well.
DIY grid-tied residential solar+inverter+battery. Trying to design the solar arrays' tilt mechanism now for lifting/lowering 5 panels at a time in winter (60-degree winter angle, 35-degree spring/summer/fall; ~24" difference). Thinking either two linear actuators, or a single hydraulic jack connected to multiple support beams. The weight isn't much, but I want a way to lift entire top edge at once to prevent twisting. Linear actuators are slightly more money and easier to build, but require power and weather-proofing. Jack is cheaper, but more complex to distribute force. Wondering if there's other options. (winch would require more robust/taller rear posts, seems more complex, might shade rear array)
Tilting them vertical or nearly so is very useful if there could be any hail, that might be a good idea to support.
What about compressed air? It might not be too hard to find a small brushless low power air pump that could drive pistons directly.
You could mount the pump controller onto the back of the panels and use an accelerometer to measure angle, and run the pump until it's where you want it.
You'd probably need to do some testing and make sure it couldn't get jammed, then build up pressure, then suddenly unstick and move unsafely.
Thanks! And I've seen Dave's tests. Since I don't need a fixed angle, do need to maximize production (due to a limited sun horizon), and only have 10 panels, the best option is adjusting angle during winter. The N-S orientation is a really poor performer. (Notice that his test is 30-degree tilt vs N-S; 60-degree tilt will provide 20-25% more power than 30-degree at my latitude, without even considering bifacials w/snow. The only thing that would produce more is an active tracker, but I've got other things to do, and a December deadline...)
Frustrated by the complexity and high overhead of most monitoring tools, I wrote Simon.
It’s a single binary, dependency-free monitor in Rust that does it all: metrics, Docker, alerts, and file browsing. While maintaining a minimal footprint for embedded systems and other constrained hardware.
I just launched a 10-Bit Video Thumbnail Provider for Windows.
Windows does not natively support rendering thumbnails for 10-bit videos, which are commonly produced by cameras like the Sony A7IV.
When I started working on a short film the video clips were piling up on my hard drive. Opening them one by one to find what I was looking for was tedious.
I could not find a reputable solution to this problem, so I started a company and built one. I went through the process of EV Certification to have the installer and executable code signed.
I hope to be in the Microsoft Store soon.
I'm also building other utilities with similar purpose.
Originally started in 2012, I’m (still) building log.soccer - a stat-tracking tool for amateur soccer players. This is the third or fourth iteration of the site, which until this year mostly served as a glorified résumé project to showcase the latest framework or tool I had just learnt.
Thanks to ChatGPT, my productivity went through the roof this year, and I finally shipped an MVP that might actually be useful.
It’s a standard Django + React + AWS stack. My raison d’être is to build an Apple Watch app that tracks match scores in real time. 2026 hopefully.
We are building end-to-end accessibility compliance tool[1] that will take care of auditing, remediation, verification and generation of ACR/VPAT.
Because of the well bound nature of the problem space, we are able to unlock a lot of power from LLMs and put together a good end-to-end product that delivers the promise.
Still early days. I know there are lot of folks who care about a11y. I would love to chat and learn from your experience.
Instead of sending the page's HTML to an LLM, Hikugen asks it to generate python code to fetch the data and enforces the generated data conforms to a Pydantic schema defined by the user. I'm using this to power yomu (https://github.com/goncharom/yomu), a personal email newsletter built from arbitrary websites.
I'm working on SuperCurate (https://getsupercurate.com), which is geared towards note retrieval and curation rather than note creation.
Think filing cabinet for your notes, web clippings, images and PDFs.
I wanted fast search and filters for my Evernote archive so I could drill down and surface exactly what I was looking for.
I’m working on a performance capture library for Python because I often need to know the performance of backend systems I maintain. I frequently build tooling to capture performance and save it for later analysis. I/O operations get costly when writing lots of data to disk and creating good real-time analytics tools takes a lot of my time. I wanted a library that captures real-time performance analytics from Python backends.
This is why I wrote kronicler to record performance metrics while being fast and simple to implement. I built my own columnar database in Rust to capture and analyze these logs.
To capture logs, `import kronicler` and add `@kronicler.capture` as a decorator to functions in Python. It will then start saving performance metrics to the custom database on disk. You can also use the middleware for FastAPI.
You can then view these performance metrics by adding a route to your server called `/logs` where you return `DB.logs()`. You can paste your hosted URL into the settings of usekronicler.com (the online dashboard) and view your data with a couple charts. View the readme or the website for more details for how to do this.
I'm still working on features like concurrency and other overall improvements. I've added a lot since the last time I shared on HN. I would love some feedback to help shape this product into something useful for you all.
We've been tinkering with building realtime talking head models (avatar models, etc.) for a while now, and finally have something that works (well enough)! Operates at ~2x realtime on a 4090, significantly faster than that on enterprise grade GPUs.
The main use case we designed for was language learning, particularly having a conversational partner -- generally we've found that adding a face to the voice really helps trigger the fight or flight response, which we've found to be the hardest part of speaking a new language with confidence.
But in building out the system around the model to enable that use case (tool use on a canvas for speaking prompts and images, memory to make conversations less stale, etc.), we think there's potential for other use cases too.
Tools to help my mental health tracking[0], and sharing with others how I manage my limited amount of energy[1]. They're kind of related, since mental health impacts my energy, so I've needed to prioritize and really make sure I'm spending my time and energy on things that matter. Usually, there's a good mix of things I enjoy doing with things I gain a lot out of. I've spent a lot of time thinking about this!
I've used my app in various forms for around 5 years, rewritten multiple times. But now I'm creating surrounding tooling to help others put my mental model for personal life prioritize to use. I'm writing in the "Saving Spoons" Substack as I go, trying to explain why and how I do things, with advice for others trying to do the same thing.
Is there a cadence for these threads? I had in mind to "be prepared" to post in November's with what I'm working on, but I expected it to come around on the 15th (mid-month).
(The proprietary apps are built with the toolkit).
I've struggled to pitch or articulate the vision here, but my latest pithy attempt is: scaling self-actualization by mechanizing the nested loops described by Anders Ericson's 'deliberate practice'
- Inner loop: individual learners maximize their skill uptake velocity and performance peak by adhering to domain specific best practices
- Outer loop: domain specific best practices get refined according to innovation or serendipitous discoveries from the inner loop (eg, someone is observed to beat out prior best practices)
As mentioned, I'm flat-foot posting here, so the pages aren't all prepped. https://flutor.app/dbg and https://letterspractice.com/dbg show some of the innards. Not linked, but I'm especially fond of https://letterspractice.com/dbg/juggling - the premise here that as child practices the letters, the letters exemplify the principles of effective practice in alliterative skill domains (juggling Js, batting Bs, flossing Fs (it's hard ok?))
Disenhackifying one of the last pieces of my KaithemAutomation server that still feels not best practicesful.
Device driver plugins used to have a very simple flat key value, strings only format, with a set_config_properties function to tell the host what kind of UI to show.
That's all getting replaced with JSON schemas, with some auto-upgrade shims so old config keeps working.
It's one of many things that now seems completely insane, but made sense when I had way less experience a long time ago!
Also still on and off working on my BLE/WiFi based Meshtastic-alike.
Loads of similar products out there, but non that did all of: open source code with attested releases, recorded mic and system audio to work with any meeting app and used Apple Intelligence for private summarisation. In beta, and also just released a experimental version with self hosted Ollama support.
Currently working on training language models steered towards certain "states of consciousness".
I have a model trained on publics datasets tied to brainwaves and/eye tracking and text comprehension (have this working well enough to experiment). Now I am training an adapter for various llm architectures to generate text steered to certain neural oscillation patterns (let's call them "states of consciousness" for brevity). I also have a 'rephraser' that rephrases text to elicit these certain states of consciousness. Overall experimenting with creating an suite of tools off my findings with how text relates to the eigenmodes of consciousness. My theory is once I do this I'll be able to do some...interesting things with "AI" agents. lmk if you want to talk about it if you're someone with knowledge in neuroscience/ML. My background is as a Software/ML Engineer so I could use additional thoughts. I do wish I could send a Github/docs which I will soon but this is currently a private project seeking investment for various research/public/private sector applications.
Curated LinkedIn topics + AI drafting: validating before building
I'm exploring building a weekly curation service for professionals who want to write on LinkedIn but struggle with "what's worth writing about."
The thesis: In the AI era, execution (writing) is commoditized. The real bottleneck is editorial judgment... knowing what topics matter before they're obvious.
The concept: Weekly email with 5-7 curated topics (tech trends, policy shifts, market movements). Each topic comes with sources, multiple angles, and context
Choose your perspective, AI drafts a polished article
Why I think this could work: I've been manually doing this for myself for years. Pattern recognition at scale is hard to automate, but pairing human curation with AI execution might work.
Target market: ~30M professionals who should be building thought leadership but don't have time to spend on research.
Current status: Validating demand before building. The hard part isn't the AI, it's systematizing the trend-spotting and curation process without losing signal quality.
Bread and butter stuff. Pulling together all of the assorted algorithms and data structures I implemented in C over the years out of necessity - lists, trees, stacks, queues, hash tables, memory pools, etc. - aligning the APIs, cleaning up and merging into a library. It's a background project but super fun. This and several parsers - JSON, some config file formats, and parsers for some GPS / GNSS receiver data protocols. FSMs also always feel like nice, clean fun. And prematurely optimising every bit.
This helped me bridge the gap between installing packages declaratively via NixOS / home-manager and defining them for each project being worked in via flake.nix / direnv / nix-direnv; which was needed since most projects don't use Nix.
I’m building Sink It for Reddit (https://gosinkit.com), a browser extension to make Reddit usable on the web. It’s similar to RES with broader support for all the different Reddit UIs (there are 4).
It’s mostly free with only old Reddit features gated behind a one time $5 fee. The app has a few hundred thousand users on the Apple platforms but recently it was invited to join Mozilla’s Recommended Extensions program so I’m hoping to grow the non-Apple user base.
A multiplayer game prototype, written in Rust, using Bevy + egui for graphics. Think of it as a bare bones implementation of a game like Runescape, mostly to test out current LLM capability.
I haven't found much value in LLMs for coding beyond very self contained tasks, but some people speak highly of it, and I want to be sure that I'm not missing out. So from time to time I give new tools a try. This time is "Claude Code on the web".
I've put in an estimated 50 hours so far. It has a client and an authoritative server. The client displays 3D graphics with some placeholder models. From the client, you can click on tiles and move to them, or click on enemies to pathfind and attack them. You can right-click on tiles or monsters to open a menu with options (attack, trade, move). There are some unit tests and a few integration tests.
Right now the issues that Claude has been unable to resolve after a few attempts are:
* Attack animations. I'm trying to get it to raise and then lower a rectangular block to simulate a sword attack. It really doesn't get it, and it's harder to write tests for compared to movement and server-client networking.
* "Entity interpolation". Rather walking entities instantly moving from tile to tile, movement should flow smoothly.
I have Claude Pro ($20/mo) which let me make a few commits per day. After a few days of that, Anthropic offered $250 in credits to promote "Claude Code on the web". The credits expire after two weeks. I'm now five days into that period and have gone through $50 in credits. It is heavily rate limited and frequently locks me out for multiple hours after only a few interactions, but it's free credits so I can't really complain.
I've been working on two game development projects for the past couple of years.
One project is for building rhythm games in multiple game engines and multiple platforms. Currently, it works in Unity, Unreal, Godot, SDL (or any C++ game engine), and MonoGame (or any C# game engine), and runs on Windows, macOS, and Linux. I'm working on adding Love2d (or any Lua game engine) and Bevy (or any Rust game engine). I have a few local prototypes of it working in Unity and Godot, but nothing public yet. Still trying to figure out what kind of game I want to make with it.
The other is a general purpose game engine in C++ with SDL. It's far enough along that I'm building games in it, but it's more of an exploration into how games are made than a replacement for Unity or Godot. I suppose it could be eventually, but I'm trying to be realistic with what it can do. One thing I'm pretty happy with regarding this engine is that one of the demo repos will automatically build to WebGL and publish to itch.io when changes are pushed.
I'm working on _prompt injection_, the problem where LLMs can't reliably distinguish between the user's instructions and untrusted content like web search results.
Finished: the 100%-vibe-coded "GPT-5 reviews all my PRs on max reasoning" GitHub app (which is shockingly effective, https://github.com/Smaug123/robocop - probably nothing new for people who already use some product like this, but I like owning my own infrastructure as far as possible, and GPT-5 and perhaps Gemini are the only models smart enough to do this so I can't take this any further).
Currently: back on "write an immediate-mode TUI framework that uses a vdom as its fundamental abstraction" (https://github.com/Smaug123/WoofWare.Zoomies), in the hope that this is the first UI framework that I don't absolutely loathe.
Next: using the TUI framework, write a debugger to inspect the internal state of my deterministic .NET runtime (https://github.com/Smaug123/WoofWare.PawPrint) and to step forward and backward in time.
Next: get the deterministic .NET runtime to a point where a property-based testing framework can identify the deadlock in some very simple buggy multithreaded code. (The framework is not yet able to run Hello World - did you know that's an incredibly complicated program in .NET? - but it can solve a few Advent of Code problems right now, can perform some limited exception handling, limited virtual method dispatch, limited casting between types. Even getting to Hello World might take a year if I'm unlucky.)
Trying to keep it simple but I can already feel some "design pressure" to think about making the DSL more complete (language) by adding features like loops and variables. Still early days!
Made a website to host a blog! Right now it's empty except for one post describing the process of setting up the blog. I plan to add more stuff once I finish this semester in college.
I'm still working on WithAudio (https://desktop.with.audio). A one time payment Text To Speech Desktop App. Because I think everything doesn't have to be a subscription.
In October I finished the PDF parser. It was a big challenge extracting PDF contect with correct paragraph breaks on user's computer locally. I'm gonna write about this soon.
Now I'm working on a web extension that talks to the app that run locally on your system so you can use WithAudio in your browser with very good performance, 100% local and private.
I'm working on Argon Chess, a deterministic chess variant with some degree of cheat resistance (hard to describe to chess engines like Fairy Stockfish) and tons of variety. A week ago, I added a way to play friends online a week ago (a Discord Activity) and a simple Play a Dumb AI feature on its website. You can also print the cards for free for offline play. https://argonchess.com/
Developing a fingerprinting method for identifying music masterings! Like Shazam but to tell what version of an album you have.
The idea being able to compare measurements to see what mastering you're really getting - because they are NOT all equal. With the remasters and stealth replacements on streaming, it seems like every other month I wake up one day and my favorite music sounds worse (or is gone...). Now I can measure it and help find what versions I really want to collect!
I may end up trying to make a fingerprint database/tool that sits in between MusicBrainz and Discogs. That way hopefully the community can standardize and quantify some of this info that only lives ad hoc in Steve Hoffman forum threads or partially on sites like https://dr.loudness-war.info
It’s been a fun 3 year project. Just launched on iOS and am in user acquisition phase. Totally new learnings here! Getting users is definitely the hard part... I can build something all day
My backhand is OK but my forehand sucks. Grip styles for standard handles usually end up favoring one side or the other. I'm making a handle shape that's easier to get the blade angle right on both sides. Hopefully a couple more iterations on the 3D printer and then I can have a functional prototype made.
I'm working on https://yap.town - an SRS based language learning app.
I would say it combines the best parts of Duolingo and Anki. Anki is great for memorizing words, but you don't see the words in the context of novel sentences. Duolingo is great for exposure to new sentences, but it's oriented around "lessons" and SRS is an afterthought. (Duolingo is also not designed for people serious about learning a language IMO, it's too easy and goes too slowly.)
Had to do quite a bit to get it to work well.
1. At first you would think that if you know all the words in a sentence, that should be enough to understand the sentence. But it doesn't work like that. For starters, words can have multiple meanings. The french word "bois" can mean "(you) drink" or "wood". You want to learn these separately. I trained an NLP model (a gemma3 finetune) that I use to understand the manner each word is used in each sentence: https://huggingface.co/collections/anchpop/lexide-nlp-models
2. Even then, what about a sentence like "you'd better not"? Even if you know the words "you" "had" "better" and "not", you still won't really get this. So I use the wiktionary "multiword terms" category for each language to get a huge list of terms like "'d better" , "you better believe it", etc, and teach these in addition to individual words. And then I only show sentences where you know all the individual words as well as all the terms.
And I'm not planning to get rich off of it haha. Right now there's no monetization at all. If lots of people use it to learn a language and avoid wasting their time on duolingo, I'll be happy
Why? I love the old arcade and game boy games, and I want to recreate them to my liking. I also love mechanical systems and space rovers, and I want ro build worlds to explore and simulate these things
For the past few months, I've been building a health tracking app called LogBuddy. I got tired of using separate apps for nutrition, workouts, weight, and period tracking. So I built LogBuddy to handle all of it with a dead simple interface. That way, all the heath data I'm tracking would be in one place.
Right now there's an Android APK available. If there's interest, I'll publish to the Play Store and build an iOS version too.
Would love to hear your feedback if you try it out!
The goal is lightweight, composable tools with clean interfaces that respect user agency and privacy, provide technical clarity, and make you a better photographer by encouraging mastery over your tools and offering new ways to approach picture making. Also broadly honoring the (almost) 2 century old history of the craft and drawing inspiration from pre-digital processes and approaches.
Got a number of updates to existing apps and new ones in the works, I’m excited for the full long term vision I have that I plan to sum up in an essay at some point.
Currently Apple platforms only but the plan is also to break out of that down the line.
Still working on my digital nomad event and workation aggregator.
But now with travel and visa guides to help remote workers become productive in Japan and South Korea ASAP and give them visa guidance if they want to stay a bit longer.
Knocker, an http knock based access service for your homelab that works at a reverse proxy or firewall level.
It's a more convenient albeit less secure alternative to VPNs like tailscale.
It's more convenient because it whitelists the enite network, and it's less secure for that reason.
https://recipin.com Recipe extraction and archiving to avoid link rot and blog spam. No tracking, no JavaScript, no AI[0], and just a dusting of CSS. Source available to run your own server if you’d like (https://github.com/bradly/recipin).
I’d like to add importing from a Pinterest account and continue adding support for all the creative implementations of the schema.org recipe format that different sites use.
[0] My partner has a bunch of handwritten family recipes, so I’m trying out an optional extract from a photo of a hand written or magazine recipe that uses AI. Not required and I may pull it out into its own service that spits out schema.org recipes. We’ll see.
I'm building Your Next Store (YNS); it's a Shopify alternative built with React and Next.js.
We provide an opinionated boilerplate tailored for tools like Claude or Codex, so designers and developers can build storefronts faster and more easily. It enforces a clear structure to start from while keeping full control over design, animations, and the overall storefront experience. It’s built on top of Stripe, with our higher-level commerce abstractions, like "add to cart", "checkout", "pay", "browse products" etc; plus a Commerce CMS so merchants can manage everything smoothly once their store is live.
If youre planning to sell something online and want a modern solution, hit me up! :)
Correct and performant way to calculate historical value of a portfolio. I want a pure function, but taking a date as input is insufficient because users can edit holdings, and securities can split.
Weighing the tradeoffs of doing this calculation server or client side. That'll be an architecture shift away from my current set of background jobs fetching state and towards something more functional and on-demand.
LINOG.ph is a live earthquake tracker for the Philippines.
The Philippines deals with thousands of earthquakes a year. Whenever the government volcanology and seismology department detects earthquakes, they post it on their official website.
When a major earthquake happens, a huge number of people try to visit the site, causing downtime for up to an hour.
LINOG.ph caches earthquake data from the official government website and the U.S. Geological Survey site, and makes them highly available to the public.
I built this after seeing friends and family donating and providing support for affected families after a major earthquake in Cebu. This was my way of helping out.
Two super typhoons have hit the Philippines in the past two weeks, so I'm also considering adding in typhoon tracking.
I'm working on my own code review app powered by local or self hosted LLMs. It started as a way to lint my own code and took off from there. It's basically like greptile or co-pilot, but has some things that they don't:
I’m working on Reflect [0], it’s a privacy-focused app for self-tracking and self-discovery. You can track metrics, run self-experiments, set goals, view correlations, visualize your data, etc.
Imo, you really should use a pure white background for the App Store screenshots instead of the current greyish background which looks kind of depressing.
I built https://forvard.org/ with Tauri + Svelte; Forvard is a accomplishment tracker where your data lives locally (think of Obsidian but for career tracking), it has a bit of smartness to summarize your accomplishments (without sending your data over the network). I'm working on fixing bugs and adding couple of features!
Frontend framework in JavaScript that requires no build step, relies on DOM and SSR and can be used to build both SPA and hybrid apps without VDOM, js templates, hydration or putting HTML (or worse, css) inside JS code. It'll also have a very sophisticated declarative state manager which makes managing state and ui transitions a breeze. It's basically anti-React.
Sure: https://code.qount25.dev/qite/qite-js
No docs yet, but I suggest you go to test/demo for examples. You can actually see them work if you run it with `node test/server.js`.
I'm working on a video / post on how to solve the 1 billion row challenge (https://github.com/gunnarmorling/1brc) and get a competitively fast result while keeping the code readable and maintainable.
So far I'm within spitting distance of the winning entries without using any unsafe code or bit twiddling tricks or custom JVMs or anything like that, and having all the concerns nicely separated and modularized.
I've been building a little toy computer and assembly language that's interpreted in python. Pretty close to the first release (and introductory blog post) and a lot of fun to build (and learn a bit more about real assembly as I go).
I am building a foundational layer for building C++ apps using Bazel.
I am working on creating a standardized set of paths and third party libraries that work seamlessly across multiple developer teams. Allowing library upgrades to happen transparently in the background. This will enable developers to focus on business specific logic and not have to worry about the intricacies of the build system and allowing to "magically" work in the background. This is allow foray into Bazel and using it as a learning exercise to master it.
I'm building one project a week for the next 25 weeks for my newsletter. First, I want interesting content for the newsletter. Second, I want to try to grow the newsletter to put out something fun and joyous. The world needs more good fun.
It's really cool, but as far as I know there's no complete C++ implementation for embedded platforms, and I still can't figure out how it actually works.
Does the gossip flooding mean every single node needs to know about every other node in the entire mesh?
I have a project vaguely inspired by this and Meshtastic that tries to make use of existing internet tech, while falling back to local links, instead of trying to replace the Internet completely.
It's very much WIP, I'm planning to get rid of all of the automatic reliable retransmit stuff and replace it with per channel end to end acknowledgment.
https://github.com/EternityForest/LazyMesh#
Building https://www.hessra.net/, an authorization system based on the Biscuit token format (decentralized, signed, and attenuable). The goal is to push beyond JWTs and Zanzibar-style policy engines by giving every machine-to-machine request its own embedded, verifiable authorization logic in a small capability token. These tokens can be delegated, restricted, and verified locally with no extra network calls required after getting the token.
Early use case is replacing API keys with identity tokens that expire, delegate, and prove possession and then can be used for easy step up to fine-grained authorization. There's some pretty interesting authorization stuff you can do, like having multiple parties sign off before a token is valid or requiring a series of micro-services sign a token for it to be valid.
I rewrote Playwright to run completely in a Chrome Extension without CDP or chrome.devtools for no practical reason at all. I started to do it like Forest Gump started running. It can't get past bot protection so pretty worthless from a browser automation point of view. [0]
What I don't understand is why the need to rewrite Playwright instead of just patching it. Playwright (or Puppeteer) has addressed every edge case that has come -- especially race conditions which are a monster to deal with -- up over the years and by the time you do the same you will have Playwright.
Why is rewriting or rebuilding Playwright from the ground up needed?
Very cool. I make a consulting business out of packaging selenium scripts into windows apps for small businesses, do you have any desire to turn this into a saleable product?
Stagehand is our open source project, but the company behind it is called Browserbase - https://browserbase.com/ where we run headless browser infrastructure as a service. So no interest at this point, Browserbase drives the revenue that funds Stagehand!
Currently working on secure file intake for Intercom. Recently spoke to a customer and turns out file intake is just a part of the bigger story, currently thinking about processing files, e-signature, client portals and so much more.
Nonoverse[1], an iOS puzzle game about nonograms (image logic puzzles).
So far all levels have been handmade pixel art. I’m now testing machine generated puzzles with random “pixels”. This is an interesting challenge because I still want levels to be solvable and fun. I recently released 15 new puzzles like this and I’m preparing a new update with more.
I'm going all in on my side project CodeBrew, a Java IDE for iPad. Currently working on OpenGL support for 3d graphics, as some schools requested the feature. Also I'm finally pitting some work into aquisition, which has turned out to be much more fun than I anticipated.
Go check it out, its free to try, with a one-time purchase full version:
Working on https://gametje.com (a Jackbox games competitor). Been working on the Android TV app lately. Will probably start creating a new game next week with acronyms similar to the old game Acrophobia from the late 90s/early 2000s.
A new web server written and server management dashboard in JavaScript that is much faster and less complicated than either Apache or NGINX and serves HTTP and WebSockets from the same port.
Porting LevelDB[1] to Seastar[2], for internal metadata storage in Redpanda[3]. Before you ask why can’t something off the shelf be used, seastar has unique constraints around its runtime and its memory allocator that means we can’t reuse an existing library.
I write almost daily article about libGDX - my most favorite code-centric game framework. There are now over 100 articles covering topics from basics to advances. I plan to post more because this is more or less a passionate project.
In the future I hope it evolves into a definitive resource for learning game development with Java and libGDX.
Scratching my own (and my employer's, but they don't know that) itch and building a knowledge management system as a nerdy way of spending evenings. I refused to learn JS for years, but turns out it's not as bad as I thought, and TS makes it really nice, plus I like (to my surprise) SolidJS' JSX interpretation quite a lot. Half vibe-coded, half breaking things and learning a lot.
I posted in this monthly thread first time in May when I launched a daily logic puzzle, Clues by Sam. Since then it's grown significantly, and I couldn't be happier!
The game has a farily simple frontend, but there is a fairly complex constraint solving algorithm as part of the puzzle making process. What makes the puzzle quite unique is that you can't "guess". You can only make guesses that are provable by logic. The algorithm ensuring this has worked flawlessly for months now (though I've manually inserted some silly mistakes once or twice).
Today's puzzle is one of the hardest to date. The difficulty resets on Mondays, and then gets harder again towards Sunday.
https://github.com/s1liconcow/skyshelve - persistent python dictionary on S3. Used this to create a durable execution layer to do some of the analytics for the above.
I finally started modding a total war game (Warhammer 3). I played the series since the very first title, Shogun and I always wanted to improve the control over units and add my custom AI to not micromanage everything, but assumed it would be too time consuming and distracting from my main work. And well, it likely would have been.
But thanks to LLMs, I finally decided to give it a go and got something basic working in a short time, hurrey for AI assisted coding!
Feels empowering to be honest. No idea if I will really implement the main ideas, that I have since a long time, but I know that I can now if I want to.
I am currently working on a fork of Alt+Tab replacement Switcheroo to show all available windows in a tabular format. Current windows in the center. Apps with many windows on the left sorted by process name (e.g. Excel or PDF windows). Pinned windows such as open emails on the right. https://github.com/coezbek/switcheroo
Ups = some local-area doulas have started sharing the baby app in a big WhatsApp group & growth is starting to pick up.
Downs = my first vibe-coding horror story. For PracticeCallAI, the subscription flow was failing and somehow outside my test coverage, so I've been missing out on new subscribers for the last two months. In an effort fix it, Replit Agent - which I have been loving otherwise - truncated the table that stores all of the user calls. and their database rollback is throwing errors. So that's been fun.
I am working on methods to automate my VC firm. We have a small team and many different tasks to do. I’ve had success with using LLMs to help us automate various projects. But I appreciate any open source tools, techniques, readings, etc. if anyone knows any!
I'm adding an overly elaborate item and levelling up system to the adventure mode of my chess variant AI sandbox: www.chesscraft.ca
Items have a prefix and suffix system similar to Diablo 2 so I'm having nostalgic fun building it. None of this gives any advantage to the chess games you play. It's just a pointless cycle of gems, items, and experience to get more gems, items and experience. Seems fun so far.
building a location-based game that captures Taiwan's absurd convenience stores density[1].
Players stand at any 7-11/FamilyMart/etc, take a photo showing you can see the next store, walk to it, repeat. Chain as many stores together as you can.
It sounds silly but Taiwan really is this convenient - you often can see 2-3 stores from one spot. Here[2] one route where you can actually link 7 convenience stores in a row! Now trying to make maps look a bit nicer with mapbox.
A Mac-based video manager that automatically transcribes, translates and summarises videos. I process information best through reading, so I built it to manage my growing collection of training course videos, webinars and meeting recordings. Currently working on adding RAG search to make it easier to query content.
Also building a CMS and static site generator that runs entirely client side in the browser. Pick themes, model content an publish to clean HTML. It also makes content available beyond just the browser, eg in a command line TUI.
I’m building a typed, array-oriented dataflow compiler that takes small declarative schemas and emits plain Ruby and JavaScript, with a C path. It has a mid-end with inlining, common subexpression elimination, constant folding, dead code elimination, loop fusion, and LICM.
On and off working on the Navigation API for Node, Bun, Deno, & as a browser polyfill.
Has 90% test coverage, makes use of web platform tests to verify compatibility, and is in use by some larger companies already with the Navigation API soon to become a baseline in evergreen browsers.
The Navigation API effectively is async state navigations. The likes of React has recently added Navigation API support to make use of the browser reload indicator.
I'm working on Travi, an AI-powered travel companion that helps travelers effortlessly discover the best attractions to check out based on their interests, and experience them through rich, immersive audio narratives.
I'm working on a web app that creates easy-to-understand stories and explainers for the sake of language learning. You can listen in your favourite podcast app, or directly on the website with illustrations.
Most of the testing so far is English/French/Japanese/Mandarin, but I'm eager to add more languages if anyone is fluent and willing to help me evaluate the text-to-speech.
Building https://ottex.ai - a native MacOS app to solve repetitive micro tasks on a computer.
- Transcribe voice to text (especially useful when you need to explain something to Claude code )
- (soon) select text to instantly Check grammar / Improve writing / change tone of text
- (soon) select text to Translate between languages
I discovered that I have a few 10/20$ subscriptions (grammarly, raycast, wisperflow) that do embarrassingly simple stuff I can one shot with cheap SLM. So I decided to build a one app specialized in small repetitive tasks on computer.
An Intent is a self-contained document that describes a user request. It is composed of three main sections: WHY (the motivation), WHAT (the requirements, often in Gherkin language), and HOW (a detailed, step-by-step implementation plan defined with tasks). This approach ensures clarity and alignment before any code is written.
Working on adding Apple Intelligence to my macOS app built to analyze iOS app size metrics. I'm hoping to have a locally running assistant that can act like an iOS build engineer to provide optimization opportunities and more: https://apps.apple.com/us/app/dotipa/id6742254881.
Right now my app allows users to export build metadata as JSON which can be interpreted by LLMs for analysis, but I'd like to have this work on-device.
Create a script for a product demo or tutorial for your app using an extension. The script is used to generate your product content in multiple formats (narrated video, interactive demo, looping animation, and in-app guide). Whenever your product changes, just update the script and regenerate everything. No manual re-recording of video, syncing of audio, or any other post-production steps.
New (open source) PostgreSQL index type for analytics workloads, which is a read-only drop-in replacement for B-trees. Smol is multiplicatively faster than B-Trees and radically smaller.
In the philosophy of selling shovels in a gold rush, I have built a Markdown Viewer for Mac which is optimised for AI coding with the likes of Claude.
It is simple but powerful supporting all formatting but also diagrams so you can get Claude to generate beautiful ER, or state-transition diagrams for your documentation. It also supports math notation, file links and has a cool table of contents feature
I'm building a scraper in Golang based on Colly to do two things:
* Automatically train the scraper on the structure of the page to acquire the data you want, and
* Clean and structure the data into a format suitable to go into a relational database
I got sick of doing all that manually for some pricing data I wanted to monitor on some suppliers sites, and I've always wanted to contribute more to open source and give back.
People use Puter for an incredibly wide range of things, including cloud storage, web hosting, coding, AI, and gaming. Right now, we're mostly focused on improving performance and making sure that it's as fast as a regular desktop environment!
This looks fun! But the fact that the self hosted version lacks support for some core apps is sad, I'd love to be able to build puter apps! Any plans for an app store-like ecosystem?
Multiplayer QWOP-like where you control one leg of an octopus.
I'm further ahead in the development than shown here, hopefully have the finished thing out with support for multiple games within a month or so (would be faster if I didn't have a job lol)
I'm working on a command-line tool for advanced full-text search of written documents. It works in a completely different way than grep, so it can do a lot of operations that grep fundamentally cannot like proximity searching.
I called it Wosp for word-oriented search and print. I released the first functional version a few days ago: https://github.com/atrettel/wosp
Eyeball is a bookmarks app that turns your own saved links into hyper-personalized playlists. It's like having a personal curator in your pocket that sends you a weekly issue of your own personal "magazine" on Sundays.
Working on AI business coach that can help you at different stages of your startup or small business journey. Here is where you can access the business coach:
https://ceo.getbeyondx.com
I'm working on a Yelp alternative called Vibehuntr -- just something different to browse venues using Google's API, with a social layer so I can see what my friends like. It's very rough around the edges right now and it might be completely different by next week. It's been a fun experiment in vibe coding on a full stack. https://vibehuntr.io
An LLM-powered 'offline' journaling/mindfulness app that draws on ancient philosophy. Designed it initially to help nudge my own habit along & keep things fresh/interesting every time I sat down for a scribble-sesh.
Some years ago, I co-founded a startup that would run workflows when email messages arrived, any email from any source was parsed and it would trigger "actions" that could include notifications - this sounds like a good use case for it! You don't need the service to expose an api to listen in, since most services end up sending email as last fallback.
This is a pet project for myself. I love listening to online radio while at work, helps me focus. But I didn't really click with any of the current selection of web apps out there so decided to build one myself.
It uses the great API available at radio-browser.info for all the radio information.
Been using it as a way to learn how to market a website as well. Learning a lot.
Would suggest that you filter out any radio stations where the URL isn't working if possible.
For example I filtered down to "United Kingdom" and then "bass" - 3 of the 6 worked and would rather see ones that are active.
Also if possible to apply the country filter within the search bar, took me a second to realise I had to open the filter for country, select that, then go back to my search.
When clearing my search of "Bass" in the example above, it reset the search to default (didn't have my country filter) even though the filter was still applied when opening the filter section.
Super easy interface to use though, really well done.
I'm working on https://teeming.ai, trying to solve the information asymmetry problem in the job market.
The project has been a huge learning curve for me - I started out as a skeptic of how generative AI could solve real problems (rather than just create noise) but now think that, like the internet, it can create a new kind of abundance that will be harnessable in all sorts of interesting ways.
Unlike traditional accounting platforms we expose the ledger model directly which enables our customers to model complex transactions even when we do not have direct support for it.
Been working on this for a month, and it uses Elixir, Phoenix and InertiaJS with React.
I have found that duplicated tabs can be useful e.g. for pages where footnotes are not hyperlinked in the text. When this happens I open a duplicate tab and scroll to the bottom of the page on it.
oh, for sure, that's why the extension shows which tabs are duplicated, and I can kill the duplicates individually, but also has a kill-all-duplicates button
Specifically working on our FinOps agent which can identify and remediate cloud infa cost related issues across AWS, Azure, Datadog, etc. The agent lives in Slack and surfaces cost savings initiatives for teams to inspect and approve for the agent to fix.
I'm building a cursor style ai agent but for planning hikes/trips. It does context management and tool calls into data sources and navigates the world to find interesting places. Should be getting out of private beta this week! https://wanderfugl.com
A FLAC encoder/decoder written in Guile scheme. I struggled to get the decoder working with most test files for a while until recently. It's more or less a fully functional decoder now. It's also 1:1 with the reference meta-flac command currently as well.
I created this recently but have let it fallow in the last month. Planning to update it over the next few days / weeks. There are a crazy number of directions I could take it.
A filmmaker community for those wanting to showcase their work. Right now everyone's got their own squarespace and the problem is that about 0 filmmakers also want to be web masters.
A database populated with audio metadata (including a link back to YouTube or Spotify or whatever) that includes vector embeddings for the audio. That way I can grab clips of music I like from YouTube, generate vectors for them, then find similar things in the database.
It's off to a rocky start though, as I've initially populated it with YouTube-8M and AudioSet, neither of which are music-specific. The search results can be... Weird.
Building a set of experiments that explores LLMs visual understanding of your photos to learn about you, especially given the recent learnings from deepseek-OCR. Part of the experiments delve into storing the memories with GraphRAG so they can be effectively retrieved without losing too information.
Hi HN, we’re a Milan-based fintech startup developing FELKO, an AI-powered data platform that helps banks and credit-holders standardize, monitor and act on debt portfolios in partnership with collection agencies!
I want to make it easier to just quickly enable wake lock on your device in a cross platform, no install, offline capable way. It's a silly little project but I'm super proud of it.
Working on https://ziva.sh/, an AI agent for game development. It uses MCP to integrate with Godot, a leading open source game engine.
It's coming together really nicely, targeting a beta release later this month. If anyone is interested in game development and wants to be a beta tester, lmk :)
A cpp code generator like esphome, to generate the firmware for midi devices in a simple yaml file, for raspberry Pico.
It would have been so much easy just to program the midi hub I wanted to program but wanted to make it generic.. now I can make the firmware for any configuration in seconds!
Helping my recent MBA grad sister make a simple python script to fit here resume to a JD using openAI's api. Shes applting to product and marketing roles in AI and this helps her understand the tech (and its limitations) better as well as apply to more jobs easier
Market is brutal though man. She hasnt gotten an offer after so much trying
My latest is Marvelogs (https://www.marvelogs.com) - always wanted to build a price tracker (or tracking any values on a regular basis) - it's nearly there.
system to test and calibrate an analog traction control system. the system uses a frequency to voltage converter and a bunch of opamps to compare wheel speeds then determines wheel slip or slide and either reduces engine power or braking.
Test system uses ADCs, DACs and a DDS to produce a sine wave that simulates wheel speed.
I wrote a pretty complicated set of GNU Makefiles for a simulation library at work, but was annoyed I had to work so hard to avoid collisions, so I'm working on a "more sanitary" build-your-own-build-system/build-system-kernel type deal.
Evals for programming languages with formal verification. It's not clear how far we are from good coding performance in less popular languages in general, and formal verification has some quirks on top also.
I'm building a Firefox, Chrome, VSCode and OpenVSX security scanner and profiler, and working on building a private web store for Enterprises to switch to rather than using the default stores given all the ransomware and malware activity in that space. Will show HN very soon!
I am working on a web app that doctors/NPs/PAs can use to automatically rewrite complicated and verbose medical progress notes. The amount of time medical providers spend on documentation is ballooning, and only a small portion that time is actually spent doing medical decision making. The rest of the time is spent incorporating (ie copy editing) data as it comes in from imaging/bloodwork/consultant advice. The goal is for the app to be:
entirely self service, without needing EHR integrations
able to persist and reuse the user's writing style, without actually saving any of the notes
No projects at the moment. Just working on myself and improving some things in my life, job, cheaper place to rent, lose weight etc. Dreaming of starting a business, I just want to add a cool service to my local city but the economics is hard
Just got a 3d printer (Bambu a1 mini) and my girlfriend brought home a whole bag of plant cuttings. Thought I would give a modular plant pot (i.e with elements that allow for expanding the pot) in fusion 360 a shot.
I'm working on MedAngle, the world's first agentic AI Super App for current and future doctors. Invite only, 100k+ users, 150m+ questions solved, tens of billions of seconds spent studying smarter.
A non-bloated HTML, CSS and pure Vanilla JS framework to create dashboards.
A cross-platform JSONL viewer where I am learning ImGUI. Haven’t found any other open source GUI framework that‘s small, provides out of the box components for tables, sorting
It's a honeypot system that uses AI to mess with attackers. When someone tries to hack your app, it detects them and serves up fake responses based on attack type.
The system learns from attackers behavior and creates convincing decoys to waste their time and frustrate their efforts. It's basically a trap that gets smarter the more attackers poke at it.
Been nerd sniped recently so am working on a Rust version of markdownlint-cli2. I'm tired of having a node dependency in my projects and this seems like a constrained enough problem space that I'll actually get around to doing it.
A template-based automation tool for small private equity firms. It has some AI functionality as well to easily parse documents and information from transcripts. Basically I want to free up investment teams from admin tasks so that they can spend more time on evaluating deals and building relationships.
A lot of the AI-powered applications for private equity firms are focusing on the multi-billion dollar firms.
I'm personally tired of getting stuck in config/deployment hell every time I want to deploy a long-lived web service. Sure I eventually learned how to use systemd, but systemd has SO many things baked into that I simply don't need. systemg is a lightweight process supervisor that features everything you'd typically want when running/managing production web services in the wild.
Moved from nodered only to a hybrid of nodered and home assistant. Added some new sensors, nfc tags, modes and automations for multiple tenants / cost savings. Its been fun to automate some boring tasks.
Is anyone working on or knows a library for evaluating LLMs for application features and/or application features that use LLMs? I am wondering what people use or if anyone has their own solution.
I'm working on some writing on my blog [1], trying to improve my writing and explore style.
My current series of post follows the surge of interest in UUIDs with the uptake of UUIDv7. I've seen some subtle misunderstandings spreading, so I dive into nuance. This has spun off some mini projects, like an RFC compliant UUIDv8 implementation based on XKCD 221 [2] (humor intended). I think I have two more in the blog series.
I am building a coding agent for small businesses. The agent runs on Linux box on own cloud. Desktop and mobile apps to chat with AI models and generate software as needed.
SSH based access with HTTP port forward. Team collaboration, multiple models, git based workflow, test deployment automation, etc.
If you're into movies or filmmaking, it's a fantastic AI tool for consistent, fully-intentional scenes with deliberate set and actor blocking.
It's also the cheapest model aggregator service out there. You can log into every AI image and video provider directly and don't have to pay me anything to use the tool. You can use your Sora account, Midjourney account, Grok account, etc. It'll soon let you log into other aggregators like OpenArt, plug in your FAL API key, etc. so you can use your credits/funds wherever they happen to live.
Unlike the other "model aggregator" websites like Higgsfield, this is a desktop app written in Rust that you can keep. It also has highly intentional 2D and 3D design surfaces especially built for design.
Text prompting sucks for artists and designers, so I'm trying to put image and video design onto canvases that you can intuitively mold like clay.
I’ve recently written ImapGoose, a daemon which keeps a remote IMAP mailbox in sync with a local tree of Maildir: https://whynothugo.nl/tags/imapgoose/
It relies on “modern” (2009) extensions to minimise traffic and avoids polling entirely (relying on the server to notify of new messages or changes as they happen).
It’s currently quite stable. The only known issue is that it can take a while to detect a timeout when the system is suspended and woken up again (there’s no portable API to detect suspend/resume).
Since then, I’ve been working on a simple TUI email client based on notmuch and maildir. So far it works really well for processing email, but lacks any capabilities for handling attachments, composing, sending (these are obviously on the roadmap).
Oh I'm super curious about your notmuch project. I've been wanting an email client that's just a tui on top of notmuch. Wonder if we can pair up? Ill ping you over email
I just finished a little webtoy. It's like a comic strip time machine - you can see a virtual newspaper comics page for any date in the last seventy years.
V happy with how the CSS came out, except I spent a lot of time on an "ink bleed" newsprint effect that (oops) only looks good on HiDPI monitors... lessons learned I suppose
I buy and operate e-commerce brands that sell on Amazon, and I'm working on handing as much of the operation of the business off to AI as possible. Doing this both for actual time savings for myself and also as my big-picture eval of new AI models + products as they come out.
I also started a Substack to document it - here's a recent post on using Gemini to screen inbound emails with prospective acquisition targets via a Google Apps Script that evaluates the listings in those emails daily: https://theautomatedoperator.substack.com/p/screening-inboun....
Ahhh this is exactly what I'm looking for! I don't see any pricing on any pages. Would love to know how much this costs (I don't know what 455 diamonds is worth) as there's a few sprites that I'd love to animate and use in my app.
Not a fan of signing up before seeing how much I'd have to pay. The examples look great though.
Prices are about to drop dramatically. Many of the models dropped >80% in price since initial launch. Any time I have a reduction in cost, I pass the savings directly on to users.
Added compression to my metal 3D printer slicer exported CAM files, and refactored the code to better support larger volumes.
Tweaking the piezoelectric driver PCB design for the micro-positing microscopy stage project. The Nanomotion piezoelectric motors were not meant to be used in the manner I chose, but it is fun to push the limits of technology.
Finishing up some custom 1U mounted hardware, and getting a batch of test PCB soon. Bend radius came back 1mm oversize, but this was acceptable for a single run item.
Also involved in several other projects maybe 3 people would care about. Doing a custom FPGA PCB is not very fun unless encountering that rare class of problem CPU/MCU simply can't handle cleanly. =3
But of course the programmer in me needs to make my own software to design patterns with code. Enjoying using paper.js to do all the complicated math to calculate lengths and angles.
All-in-one router/nas/firewall/adblock/app server (each piece optional)
Declarative and reproduceable as it is built off of NixOS, but administered through a UI, so the user doesn't have to know this.
All state managed in a backup bundle, so it can be hosted at home or in the cloud.
Goal is to have a box you plug just like a wifi access point into your modem, follow a simple web-based installation flow, then you are running a personal cloud.
Website is self-hosted by HomeFree, but installation instructions are very out of date, which I'm working on right now. There are now installation ISOs that I will soon add a link to.
I've been building SageNet, a voice-first AI coach that turns your goals into structured, adaptive learning plans.
After a 2-minute voice conversation, Sage generates a personalized 6-module roadmap with build-first projects. It checks in by voice, analyzes your reflections, and regenerates your plan if needed. You can invite friends to your Support Squad for accountability.
The biggest insight so far is people don’t want “infinite content.” They want structure and someone who remembers them.
In network code: most people just let the OS choose a default adapter. It works fine, but it makes it hard to write software that works across machines with either (1) multiple NICs (and/or networks they point to.) or (2) multiple external Internet IPs. Look at STUN, for example.
A STUN server that lets people test what type of NAT they have uses two IPs. For such a server you have to manually specify the addresses to bind on to make for sure its setup right. As it goes, writing network software to do simple things like "bind on all local addresses", "bind publicly", "bind on all", is harder than it sounds. There are edge cases on different OSes and address families, so manually managing IPs is hard to do.
My network software lets devs easily manage NICs and routes they support without guessing about addressing. Additionally, I've written a bunch of software with the library already to do things like NAT traversal. So its really my own redesign of how to do networking on the Internet. Designed to hide a lot of the messiness. I'm still improving code quality so it's not ready yet. But I've been dog fooding with a lot of software written in it and smashing bugs every day.
For fun have been creating a mashup of old school DnD map generation using Commodore "10 Print Chr$(205.5+Rnd(1)); : Goto 10" style logic (in TS/Svelte/SVG):
Have been down a rabbit hole ensuring the stairs are realistic and that grid connects properly. Lots of fun and frustration with AI coding tools trying to solve that (they mostly don't/can't). Some fun detours learning a little Prolog to help out as well.
I recently added FSRS (besides also having Anki integration). Now I'm working on replacing the need for reviewing flashcards by having reading activity automatically mark flashcards (current and future) as reviewed, so that you can get many of your reviews in just by reading native materials that interest you instead of sacrificing most of your study time to contextless flashcard grind.
I'm also working on a manga mode using a new manga OCR tech I have licensed out of academia that is ahead of state of the art alternatives.
This looks super cool. I love to see people working on language learning tech. I'm working on a language learning app too but it only really works well for indo-european languages. That said I would still love to collab or talk shop, my contact info is in my bio
ELO translated to the NFL with margin-of-victory adjustments, a modest home-field term, and week-to-week recency weighting.
Post-hoc calibration with isotonic regression so 70% predictions land near 0.70 empirically.
Monte Carlo to roll games forward for distributions on weekly win odds and season outcomes, plus basic reliability/Brier/log-loss tracking.
# Where I’m taking it (ensemble ideas)
Blend a few complementary signals: (1) pure ELO strength; (2) schedule-adjusted EPA/Success Rate features; (3) injury/QB continuity and rest/travel effects; (4) a small “market prior” from closing lines; (5) weather/play style pace features.
Combine via a simple stacked model (regularized logistic, isotonic on top), or a Bayesian hierarchical model that lets team effects evolve with partial pooling.
Separate models for win prob vs. expected margin, then reconcile with a consistent link so the two don’t disagree.
Emphasis on calibration over leaderboard-chasing: reliability diagrams, ECE, PIT histograms, and backtests that penalize regime drift.
# Why I’m doing it
It’s a sandbox to teach myself Monte Carlo and ELO end-to-end—data ingest → feature plumbing → simulation → calibration → eval—on a domain with immediate feedback every week.
# How this connects to my day job (healthcare ops)
I work at BlueSprig, running ~150 ABA therapy clinics. I’m exploring whether ELO-like ideas can augment ops decisions:
“Strength” ratings for clinics, care teams, or scheduling templates based on outcome deltas and throughput (margin-of-victory ≈ effect size/efficiency).
Monte Carlo for expansion planning (new-site ramp curves), capacity/OT forecasting, and risk-adjusted outcome monitoring with calibration so probabilities mean something.
Guardrails for fairness and interpretability so ratings don’t become blunt scorecards.
# Help
If you’ve shipped calibrated ensembles in sports or have pointers on applying rating systems to multi-site healthcare operations, I’d love to trade notes or if you need someone to this and other kind of work for their dayjob email me at mgracepellon@gmail.com -- I would love to do this fulltime.
Note that I am trying to narrow down a bug in my backend which sometimes causes it to crash. Since backend is built in Swift using SQLite as database, it's a bit hard to nail down the issue.
And an iOS expense tracker focused for frequent travelers, and macOS photos viewer based on the filesystem instead of a monolithic opaque "library", 2 needs that I had since forever but could never get through Apple's atrocious developer documentation far enough to finish making them :')
I've been wanting to learn more embedded type projects, and I've been snacking too often so I've been building a box that will only open on the weekends.
I got all the components, tested it on a breadboard, learned to solder and now I'm working on the 3d Print to enclose everything.
I actually just did a test run to see if my current 3d design would fit my PICO board, and it fit, but not that secure yet.
Im a developer but never worked this close to metal, so I've been so happy with how it's been going so far, making me real proud of myself.
Great project, hope you enjoy (not) using it!
I recently launched a daily word puzzle!
https://tiledwords.com
It’s inspired by tile placement board games like Patchwork and crosswords. You rotate and move tiles to rebuild a broken crossword.
It’s free, web based, and responsive.
I currently have several hundred daily players and growing. My wife and I create the puzzles and I’m continuing to fix bugs and add new features.
I just launched a ”community puzzle” feature to let players help build new puzzles.
I’d love to know what you think!
I saw your Show HN post a few weeks ago! Really appreciate the smoothness of your UI and the simplicity of your onboarding, I see how much you have dialed in. I've been working on a daily puzzle game too (it's getting there...), maybe you'd enjoy it https://slab17.com/
You had me at Patchwork. This is super fun. Thanks for making it!
This puzzle is genius.The interface is minimal and user-friendly, everything feels smooth and intuitive.
Wow that is a clean and responsive interface! It feels great on mobile.
This is really fun — have you played with making the tile position opinionated (not agnostic)?
i wonder if have the clues point to a starting square (e.g., "E5") would be better than the current "reveal" aid. The spatial information would become more helpful toward the end when the player is dealing with the words they need help on.
Could you expand on what you mean about opinionated vs agnostic? It sounds interesting but I’m not sure I follow.
I like that clue idea! I want to change how the reveals work. I’ll play with that!
I love it. I struggle more than I want to admit, but super fun nonetheless.
Great game! The effort you put into animations and interactivity really pays off, especially when first learning how the game works.
This is a classic HN comment but I’d love a Thursday/Friday crossword difficulty equivalent in addition to the dailies which are a ~Monday.
The animations in the interface make it feel more "jelly" and not "wooden" like a number of other such interfaces.
Thanks! I spent a long time trying to make the core controls feel intuitive and natural to use
The amount of care you put into it must be massive; a noticed so many nice subtle details that make interacting with the pieces easy and fun. Kudos!
Awesome work
Awesome game! I've been looking for something like this.
Thanks, I’m glad you like it!
That was wonderful, I'll be back tomorrow.
Thanks, I’m glad you like it!
this is very cool, noticed vue and nuxt nice.
Thanks! Yeah I love Vue and Nuxt. They worked great for this project
That was fun, I’m in!
Awesome!
There are many language-learning apps, but almost none that focus on improving conversational Hindi for kids.
Made this web app for my nephew, based in Singapore, after watching him struggle to find anyone to practice Hindi with outside of family calls (since most of his friends are Chinese). The idea is to have a 24x7 partner to speak with Hindi and make it fun. This can complement the formal Hindi classes that most kids of Indian diaspora parents take.
My nephew started using this and is enjoying it!
Link: https://www.hindispeakingtutor.in/
Looking to hear feedback from the community!
For the past 2 months I have been doing a heavy deep dive into image generation and image generation editing capabilities. This then had me discover that you can generate storyboards for short stories, and automate the creation of these as videos with video generation models. This is a topic that interests me heavily, and as such I am now building my own workflows around that. I am documenting the entire journey here:
https://edwin.genego.io/blog
https://edwin.genego.io/blog/lpa-studio
https://edwin.genego.io/blog/ai-branding
It's not something I am looking to commercialize, but I actually did drop out of film school (with semesters in creative storytelling) to pursue software 15 years ago. And I feel like this will open up a whole new way of visual storytelling as well as personal and product branding. I have gotten quite some emails about it, from interesting people in different industries, as some more strongly worded (not so nice) emails from someone in the VFX industry since I started. Its by far one of the most interesting tangents I have ever went on.
Currently working on a take on Pokémon GO + Pokémon Snap but for birding. The goal is to explore your neighborhood, find birds, take good photos of them all. Next month, I'll be doing an event to find a rare bird, excited to see how it goes!
It's still a small closed alpha, if anyone is interested: https://testers.birdlego.com
Here is a rough trailer of it: https://www.youtube.com/watch?v=yVpR8aafFjI
Oh I was hoping for this but with real birds in my neighborhood. Still neat!
I got super excited thinking this was for real birds. I would love someone to gamify birding.
What’s missing from eBird?
Certainly gamified a bit as I learned from:
LISTERS: A Glimpse Into Extreme Birdwatching
https://youtube.com/watch?v=zl-wAqplQAo
1.9m view | 2mo ago | 2hr long (buckle in, documentary by a couple young goofball brothers)
I want something more gamified. I want to go on missions or complete challenges or something. Not just identity and log stuff.
OK, I would pay for this. Definitely following!
What a good idea, sounds fun!
I would pay for this. 100%.
I built a chrome extension (with over 600,000 downloads) that lets you chat with page, draft emails and messages, fix grammar, translate, summarize page, etc.. You can use models from OpenAI, Google, and Anthropic.
Yes, you can use your own API key as well.
https://jetwriter.ai
Feedbacks are welcome.
Built a local-first Kanban board with Tauri (Rust + Svelte) after getting frustrated with SaaS tools and basic offline options. Stores data in JSON files you control, full keyboard-first UX, parent/child tasks, release management, and it's blazingly fast with localStorage + background sync. No telemetry, purely local. Curious what others prioritize in personal task tools. Seems like there's a gap between "todo.txt" simplicity and Jira complexity.
Cool. Love Tauri. So the whole board dataset is stored in localStorage? If you get to a point where the size limitations or synchronous blocking operations are an issue might consider using IndexedDB. There is a nice higher level wrapper around it called Dexie that has full TS typing support and a nice async API. https://dexie.org/
I’d love even a screenshot! Neat.
Building https://pricetracker.wtf but life got in the way lately.
Now looking to migrate bits and pieces to pg_lake from hydra/citus columnar.
FreeBSD 15.0. Rather depressingly, almost everything I said two months ago is still true: https://news.ycombinator.com/item?id=45419134
But I'm hoping to have it out by the first week of December.
A tool for creating custom Tailwind-style color palettes for web and UI design that pass WCAG contrast requirements:
https://www.inclusivecolors.com/
The interface is optimized to let you quickly explore and tweak multiple tints/shades at once so you can customize all colors exactly how you want e.g. try dragging vertically through the saturation curve in one motion to edit all the tints/shades at once, or shift whole curves horizontally by dragging between the dots on a curve.
It uses the HSLuv color space, where (unlike say HSL) the WCAG contrast stays the same when you change the hue and saturation sliders. This makes it much easier to explore accessible colors choices as you know only changes to the lightness slider will impact the contrast. You can also switch from the WCAG2 contrast checker to using APCA, which is meant to correct for inaccuracies in WCAG2, such as it being too forgiving for dark mode color combos.
Note the mobile version is more of a preview and the desktop version has more features.
I probably need to add something like a tutorial as there's a lot going on, but I've added more hints and tooltips recently. Open to feedback on what's initially confusing and what changes might help!
I passed this on to some accessibility folks at a couple conferences in the last month - everyone was impressed :)
Just launched a startup/life style business where I use AI to help people practice for upcoming interviews - https://hiredcoach.ai
Already have been told by some users that the interview prep they got from it has correctly predicted several of the actual interview questions they got, crediting its prep for their breezing through the interview rounds.
I'm really hoping it helps a lot of people!
I created a free collection of 4,300+ real website designs (screenshots, fonts, colors, live links)
https://fontofweb.com/u/fontofweb
I was tired of inspiration sites like Dribbble full of polished mockups that aren't practical. Or awwward like sites that don't represent the mundanity of most websites.
So, I spent a while building a tool that captures website design snippets. It's now a collection of 4,363 designs from 544 different domains.
For every design, it extracts:
The exact fonts used on the page (so far 561 unique font families I've found)
The precise color palette
A direct link to the live site
You can check out the full free collection here: https://fontofweb.com/u/fontofweb
I've built a self-hosted reddit-like community platform in Go: https://baklab.app
Users can create their own sub-communities, and within them, set up different categories and boards. Posts can be voted on, and board types can include regular posts, Q&A, or live chat. It's like a hybrid of Reddit and Discord but leans more towards a traditional web community. It also supports server-side rendering, making it SEO-friendly. This project is an extension of my previous Hacker News clone, dizkaz (https://news.ycombinator.com/item?id=43885998).
Continuing with my retro productivity software blog, Stone Tools: https://stonetools.ghost.io
I was getting a little bored of retrocomputing discourse being so centered on gaming, so I'm exploring the productivity software of the 8/16-bit era. I put real effort into learning and using the programs, giving my light-hearted but heartfelt assessment of its form and function for both its time and today.
Using the software inevitably gets me thinking about other things, and I explore those threads as well. For example, "Superbase on the C64" also discusses the legacy and promise of "the paperless office." A couple of other posts got some nice traction here on HN, notably "Deluxe Paint on the Amiga" and "VisiCalc on the Apple 2".
I'm hoping to build a strong monthly readership, so I'm putting in the work. It's been up for two months and five posts now, with a new one coming at the end of this week.
This is a neat project! I read the last post and I’ll work my way back through them.
Thanks, I hope you enjoy the series!
I've been enjoying the breadth of projects made possible with AI, I've cataloged over 200 of them created in 2025 here: https://jonathanclark.com/posts/ai-coding-million-lines-2025...
A few of my recent favorites: - swim lap counter in html/JS that uses the camera to watch you swim and count laps/timing - video recorder that records your window/desktop and uploads a file to S3 - video conferencing app that allows a 2 year to click on a family member face and initiates a video conference using webRTC, STUN, and browser audio/video capture with automatic bandwidth adjustments (works on all platforms with pure HTML/JS). - CUDA based ray tracer with HTML UI that can trace over 2m rays per second on my laptop for scientific study, allowing real-time display of optical parts. - chat front-end for image models like gemini-pro and openai that take other images and text as references and generate a big library of options to chose from in seconds, I've been using photoshop for decades but I tend to use this more now.
> CUDA based ray tracer with HTML UI
I'm curious if you mean they're running a raytracer on the back end, and you interact with an HTML UI, or if it runs browserside, maybe via WASM. AFAIK CUDA isn't directly compilable to WASM (yet?)
I have a node middleman that proxies request from an HTML/JS front end to a native cuda process using web sockets. To support multiple windows, the node process process provides communication between two browser windows. This lets me have render a model using 3JS in one window and a ray traced version in another window.
AI powered financial data PDF extractor sounds interesting
Discovered in-door bouldering / rock climbing and now go 3x a week, am absolutely loving it! Because of that, I haven't really worked on any side projects in a while. Perhaps I don't need to? My job advances me plenty in my field, but it is a bit of a bitter-sweet feeling in a sense, like maybe I should try to squeeze more out of my free time somehow.
I climb a lot! (Actually currently sitting on Big Sur ledge on el cap posting this). It cuts into my free time programming for sure, but imo super worth it! Enjoy it, it’s a wonderful hobby.
I’m replying from the cold east coast (from the edge of a wood chair in a lovely iykyk type of restaurant) to a human posting from el cap on hn; We have achieved peak technology. Oh yeah, I’m working on urban logistics, powered by AI.
I’ve been hesitant for fear of injury harming the ability to type, but might give it a go in the spring. Thanks for mentioning this I’m inspired to try it finally.
Couple things to avoid finger injuries: go easy on one- and two-finger pockets, use an open crimp whenever possible (all finger joints are bent the normal direction, and your palm/thumb aren't really involved), and don't bother with the hangboard or campus board for the first ~year.
I wouldn't worry about it too much though - almost all of the people I know with finger injuries were trying to push into really being competitive climbers, not just doing it casually for fun/fitness.
Oh also to keep from tearing your skin don't climb tired. (That won't keep you from typing, it's just painful.)
I'd like to add to this that do not make any food with chilli peppers like habanero or such if you just came from the gym with torn skin. I found out the hard way.
I struggled with hand and wrist pain for years from spending too much time at a computer. I did physiotherapy for years and while it helped me manage pain, I was never able to truly build enough strength to get ahead of it until I started bouldering. I took it very slowly—I spent months on very easy problems—but because it was so much fun, I kept going back. Initially, I would only go on Saturday mornings, so I had the full weekend to recover before jumping back into the work week on Monday. After a two or three months of that, I was able to climb anytime I wished. I'm still not a particularly advanced climber, and I typically only go once per week, but I am still slowly progressing, and I absolutely love it.
I’ve been climbing for 20 years and it’s the thing that prevents RSI for me and makes it possible to use a computer too much :). Certainly possible to injure fingers but would be a very rare climbing injury that would threaten coding.
Climbing easy routes in a gym is pretty low impact. It’s only when you start to move into really hard crimps or slopers where you’ll hurt yourself. I was a climber bum for years and have climbed crazy stuff around the world and never hurt myself to where I couldn’t type. A lot of bloody tape, but still able to type.
Try top rope climbing! Bouldering is injury prone because every fall is a ground fall. With top rope climbing you should never hit the ground so way less injury prone.
I've been working on a 3D voxel-based game engine for like 10 years in my spare time. At this point it's getting pretty close to being shadertoy for voxels.
https://github.com/scallyw4g/bonsai
I recently ported the terrain generators to the GPU, and increased the visible volume to 1 billion voxels cubed. I did a short YouTube video about it.
https://www.youtube.com/watch?v=bLfgjWsM1PI
I also wrote a metaprogramming language which generates a lot of the editor UI for the engine. It's a bespoke C parser that supports a small subset of C++, which is exposed to the user through a 'scripting-like' language you embed directly in your source files. I wrote it as a replacement for C++ templates and in my completely unbiased opinion it is WAY better.
https://github.com/scallyw4g/poof
Working on Spine AI, a visual workspace to think across multiple AI models.
You can chat, branch, and connect 300+ models on an infinite canvas: useful when you need to explore tradeoffs, check blind spots, or generate assets (research, slides, prompts, images) from the same board.
Try it without signup: https://app.getspine.ai/guest
I have been working on Buckaroo - my table display library for dataframes in notebook environments. Buckaroo adds table and analytics features like histograms, summary stats, sorting, and search to every dataframe. Recently I have been working to make it work better with large datasets.
This involves making it lazy for polars, allowing it to read arbitrarily large files no longer requiring loading the entire dataframe into memory. When a large dataframe initially displays, no summary stats will be available. Summary stats are computed in the background in groups of columns. Then results are cached per column. To accomplish this I wrote a polars plugin in rust that computes hashes of columns. Dealing with large data like this is tricky, operations sometimes crash, sometimes take all available memory, and sometimes they just run for a very long time. I have also been building an execution framework for Buckaroo. It uses multiprocessing based timeouts, and the caching to execute summary stats in the background.
Being able to control the execution, recover from timeouts, crashes and memory exhaustion opens up some interesting debugging tools. I have written methods that take arbitrary groups of polars expressions and produce a minimal reproduction test case through a git-bisect like process.
All of this assures that if individual columns of a dataframe fits into memory, summary stats will be computed for the entire dataframe in the background. And because it is cached, the next time you open the same dataframe, the stats will be display instantly. When exploring data I do this in an adhoc way manually (splitting up a dataframe by columns and rows), but it is error prone. This should all be automatic.
I will be presenting this at PyData Boston in December.
The Column's the limit: interactive exploration of larger than memory data sets in a notebook with Polars and Buckaroo
Working on https://tinythoughts.app, a self-messaging app
Today, I am implementing tag-triggered webhooks so they can get triggered based on the tags assigned to messages.
Use case: for example, I want to auto-tweet all my advice messages and auto-create linear tasks for `company` todos, auto-start a cursor agent when I tag messages with `Cursor task`
For some reason, I got 16 users out of nowhere, so I added a landing page last week. The connector marketplace is next, so that I can share these sub-connectors with the other users. Need a couple more weekends.
Over the years, I've read countless books. I started documenting one idea that shaped my thinking from each of these books. This idea may or may not be the core theme of the book.
Hope to document 100 ideas. Wish me luck.
https://www.jjude.com/100-ideas-from-books/
A Civil 3D plugin (Genabler) that will include all the network catalogs and collate the Civil 3D styles for civil engineers to use. There are some out-of-the-box catalogs and styles shipped with the default installation, but they are quite limited and fairly well hidden—which is not surprising, given that Civil 3D is a huge beast. As a result, they are not commonly used.
When people think about Civil 3D, they often assume it requires BIM modelers (in a sense, just glorified drafters) to create all the necessary catalogs and styles, and to assist with their use.
My Civil 3D plugin will:
1. Make standard, market-compliant catalogs and polished styles available to engineers at large. Think of it as the WordPress theme provider equivalent.
2. Make the entire process easy and painless through the plugin, with prominent buttons for quick access.
If the plugin is done well, there will be less need for BIM modelers, since for a fee, engineers could simply purchase catalogs and styles that are so easy to use they require no technical training.
As a side benefit, I also get to explore how LLMs can help me write code. It has been a while since I last updated my AI usage policy [0], and I look forward to revisiting it.
[0]: https://civilwhiz.com/my-ai-usage-policy/
Working on https://outcrop.app, a knowledge base for software teams with instant search, realtime collaboration, and LLM-driven workflows. It's built using Rust and friends. I'm looking for more early testers! :)
https://whenever.world/
It's an explorable database of films, TV shows, books and board games based around their historical setting: where and when the thing is set. It's been incredibly complex and interesting getting the (messy) data, making sense of it and trying to design a UI to explore it.
https://stevebennett.me/2025/08/26/whenever-exploring-times-...
It would be interesting to see for movies and tv series where it’s set vs where it’s actually shot
Yeah, that's a whole different thing. I saw this map once that showed places in California and how they were used to film various locations around the world. Turns out LA is very well situated.
I’ve been working on https://canine.sh which is a free, open source Heroku alternative for 2 years now.
It’s exactly the product I wish I had when I started my previous company. Running on PaaS is incredible for devex but the pricing is bonkers, and the vendor lock in makes it really hard to deal with annual price increases. We spent close to 400k / year for just 128GB combined fleet in our last startup on Heroku.
Canine tries to get the best of both worlds: developer friendly PaaS with no lockin or price gouging.
Just added build packs as a build option recently.
Also got a sponsorship from the portainer folks which lets me work on this close to full time
Hoping this saves someone the headache I had two years ago.
Hey Chris, would love to chat about this - could you email me? Cheers
Running OpenStreetMap off the grid (self-hosted to say the least) on a Raspberry Pi 500 (and to some extent a Pi Zero 2W) for Internet In a Box:
https://www.linkedin.com/posts/danielkrol_openstreetmap-acti...
All of the street and satellite tiles are thanks to maps.black. The search uses Nominatim's sqlite3 mode. I was told that it's experimental only because it hasn't been tried in production yet, so I'm sort of testing it in the process. So far I'm only doing administrative boundaries and natural features, but so far so good! I'm going to slowly add a few more types of POIs, I just don't want the database file to get too big.
Note that Internet in a Box has an OSM offering already, but the data is five years old and the tech makes it harder to update. As of today, there are much easier options on the table, and we get cool stuff like 3d buildings. Also, the search was much more limited.
* https://internet-in-a-box.org/
* https://maps.black/
* https://nominatim.org/release-docs/latest/customize/SQLite/
I'm building a coding agent, named VT Code [0]. VT Code is a Rust-based terminal coding agent with semantic code intelligence via Tree-sitter. Supports multiple LLM providers with automatic failover and efficient context management. Support OpenAI, Anthropic, Google Gemini, xAI, DeepSeek, OpenRouter, Z.AI, Moonshot AI, MiniMax, and Ollama (local & Cloud). Agent Client Protocol and Model Context Protocol fully support. VT Code supports a rich set of configuration options, with preferences stored in vtcode.toml. Has both Visual Studio Code and Open VSX extensions so that you can install in VS Code or Cursor, Windsurf, Eclipse.
I've been building it for several months now and enjoy the learning process, I also wrote a blog post and learnt a ton about terminal, ANSI processing. The learning has been immense for me, I now have working knowledge of ANSI escape codes, grapheme clusters, terminal emulators, Unicode normalization, VT protocols, PTY sessions, and filesystem operations, all the low-level details I would have never think about until I were implementing them. [1]
[0] https://github.com/vinhnx/vtcode [0.1] https://deepwiki.com/vinhnx/vtcode [1] https://buymeacoffee.com/vinhnx/vt-code
Building https://floxtop.com/, a Mac app that organizes your files and images.
It looks inside each file to see what it’s about, then moves it to the right folder with a single click. Everything happens on your Mac, so nothing leaves your computer. No clouds, no servers.
It already works with images, Office (Word, Excel, PowerPoint) PDFs, ePubs, text, Markdown, and many other file types (30+) in English. Next I’m adding multi-language support.
If you have messy folders anywhere on your Mac, Floxtop can help.
Try it for free - requires macOS 14+ with Apple Silicon: https://github.com/taranntell/fallinorg/releases/download/1....
It's cool to see this. I once saw an X thread on this and hacked a dirty tool for it: https://x.com/priyankc/status/1893112673434222985
https://pond.sh
Building a simple service to share content and simple sites in free time. Recently implemented sso with google. Would love some feedback.
Raptor - a new (Free software) way to build things like:
* Disk images
* Liveboot isos
* Container images (docker/podman)
Many build products are supported, with more on the way:
https://chrivers.github.io/raptor/builders/index.html#compat...
It uses a syntax that is inspired by Docker, but significantly enhanced.
Take a look at:
* The project: https://github.com/chrivers/raptor/
* The book: https://chrivers.github.io/raptor/
Been building an agentic movie database. https://memovee.com already in private testing.
The code for the agent is here https://github.com/upmaru/memovee-tama
Porting Ruby to Fil-C
It's very interesting because the Ruby codebase uses a `typedef uintptr_t VALUE` type to mean any of the following:
- A pointer to the heap
- A Ruby tagged value (which may be a pointer to the heap)
- Any integer value that fits in `uintptr_t`
Fil-C doesn't allow you to carry around pointers using integers, in the sense that when you do that, the pointers lose their capabilities.
But in Ruby's case, it's not as simple as changing the typedef to a pointer type, since `VALUE` variables often end up being used for integer math (including bit math, shifts, etc).
So, it's going to take a nontrivial patch to Ruby to get it to work in Fil-C. I think I'm about 70% of the way through (I started Friday afternoon).
I recreated a little tool to simultaneously mount all the commits in a git repository as directories at the same time (but re-use the same inodes for the same content).
The code is at https://github.com/matthiasgoergens/git-snap-fs
The original was in Python and actually had a decent excuse for existing for a very specific problem at work a few years ago. The new version is in Rust and exists just for fun.
This was also a small experiment in coding with OpenAI's codex. I wrote the Python original by hand---like a caveman. Codex was mostly ok at the actual code, especially once I told it to make `cargo clippy` happy, but it needed lots of help with the design. It kept insisting on extra complications and state.
But perhaps I'm a bit unfair here, because I only figured out the nice and simple design after reflecting on the connection between Linux's fuse and git's design for a while when writing the original. So it's only fair that the computer would also need some help to see how to match them up nicely.
Reverse image search to match dirty XTC tablets to lab reports https://pillscanner.app/
https://kauwenofspauwen.be/en Belgian food hygiene rating from official gov reports
Building https://typequicker.com
I’ve always wanted a typing application that’s both more than typing random words and is data-focused so I built this.
The more you type, the more the analytics system learns about your typing patterns and generates natural text to target those weakpoints (SmartPractice mode).
There’s a lot of variety as well; you can practice typing code in any programming language, or type text of various topics, use custom text, etc).
A 68030 based computer - https://github.com/jeffsponaugle/roscoe
It has been a super fun experience so far - I'm using CPLDs instead of an FPGA which makes the logic a bit more era period. I have a working system now with the math coprocessor, SRAM, DRAM, and other device support.
I am just about ready to get the VGA card I designed produced so I can work on debugging the design.
While this is fundamentally a system that ss less powerful than my apple watch, it is just fun to work on. Going back to very first principles debugging, building tools, and of course getting to exercise an old logic analyzer!
I'm working on an open source library in golang to get in depth system information from macOS.
Here's my work so far: https://github.com/BinSquare/powermetrics-go
https://easel.games
I'm making Easel, a 2D game programming language designed to match how humans, not computers, think about game logic. It also has automatic multiplayer. I've been working on it for 3 years!
Easel feels like a declarative programming language even though it is imperative, because lots of useful game-oriented features are first class. Like behaviours - you just say `on Pointer { ... }` and you have a concurrently-executing coroutine that's lifetime is managed. But you don't think about any of that complexity, you just think of your entity as having a behaviour and go forth and make your game.
It also happens to have automatic multiplayer. Normally with multiplayer you have to worry about doing everything in a "multiplayer safe" way (i.e. be deterministic and only modify the things your side has authority over). My idea was to put all the multiplayer stuff in the programming language itself, underneath all your lines of code. This way, anything you write in that programming language can just be made multiplayer, automatically. So you can just pretend all your players are in one shared world, like a singleplayer game, and the engine does all the multiplayer for you. It was really difficult to make but it makes multiplayer so easy for you now.
Easel is my idea of how games should be made, or at least as close to the idea as I can achieve with 3 years of work, and I would love for more people to try it out.
This is really cool. Nice project!
I tried doing something much more rudimentary before. Will be following
Oh, thank you!
I would love to hear more about what you were trying to do with your project before. Was it more similar to the declarative coding part, the automatic multiplayer part, or something else? Part of why I'm doing this is to explore the design space of how games should be made and I'm interested to hear what problems, issues, pet peeves, "bugbears" etc that other people think are worth solving.
It's been a while. But I believe what caused me the most headache while trying to build something like this was handling the interactions between different elements. Declaring which objects were affected by "attacks" or could be "player interactive" or "affected by player but not by NPC". Really this boiled down to proper inheritance. But I found myself so deep and tangled a fresh reset would have been better. Then determining if the object itself or an "objective manager" should perform the calculation each cycle.. etc
It was messy. I ended up having NPC, Item, Attack classes and for each a NPC Manager, Item Manager, and Attack Manager to calculate all their interactions and states.
That's why your project seems interesting because it seems to handle the heavy lifting of behaviors and "behind the scenes".
I’m working on https://regularly.co/ - A website made for inquisitive minds to get their daily puzzle fix. Still very much a WIP (mainly working on tuning the difficulty of puzzles to make it enjoyable for most). That being said I really do enjoy the unique combination of puzzles when I do them each day. I’m looking for feedback so if you do take a look please do let me know your thoughts!
Incredible. Thank you for sharing this, I love puzzles and like setting aside some time in the morning to do them. This will enhance that habit so much! A whole slew of daily puzzles, I'll let you know how it goes!
I don't understand why I can't place two kings in "adjacent diagonal" position when they're in two different regions. Something like:
..k..
.k...
Rules state they must be in different regions, row and column. No mention of diagonal or adjacency.
I agree, I had the same problem.
Personalized audio streams for language learners. Ideal for during driving or while doing chores.
https://listen.longyan.io/
At the intermediate level lots of learners struggle to find suitable content that matches their level and interests, more than a few learners turn to notebookLM podcasts to provide that, but that's a bit of a hassle to set up. So I built a platform that generates and manages infinite and shareable streams around your interests or specific vocabulary. It also provides live interactive transcripts (karaoke / teleprompter style) if you need it.
Core features work but still rough around the edges. Happy to help you out with any issues you encounter, languages to add, feature requests etc...
I suspect AI company want improved efficiencies and developing a framework that can be applied in determining the minimal-energy, maximal-efficiency architecture for ai models. Calculating the precise limits, like a Cognitive Event Horizon, where a model becomes so complicated it literally costs more energy to run than the knowledge it provides, and the Semantic Horizon, where it simply gets too complex to be accurate, etc. Lots of cool implications such as around a fundamental mathematical maximum learning rate which results in trying to get anywhere close to that that by doing stuff like aggressively filtering of the data.
I'm building a small rural ISP and web hosting service, as a way to learn about low-level networking stuff. I've got an ASN + IP space, and am working out the details with a colo, local fiber provider, and some upstreams. Right now I'm configuring the hardware itself (server, router, switch, etc) and learning all the bits and bobs (Proxmox, BGP, OPNsense, IXPs, etc)
All I can say is good luck. We spun up a co-op isp to take advantage of fiber grants for rural areas about a decade ago.
Maybe it was because of the grants, but it was a fucking nightmare getting off the ground even though we had nearly 90% of the population in three counties on board for the co-op. The red tape and regulations (in our state at least) made it clear that government runs for urban and suburban interests and actively undermines rural needs. I'm talking government in bed with large providers who had exclusive rights to run "high-speed" Internet to our towns and farms, even though they had never and were never planning on anything above dsl for most people and cable for the ones in town.
If I was more charismatic (and wasn't 1000% sure there were pictures of me doing drugs when I was in college), I would consider a run for state office, because it's a shit show for small towns here.
And that's the story of a) when we got sued by a large provider that I hope goes out of business and burns to the ground, and b) the last time I volunteered on a large project and why I will never take the lead on anything bigger than the Lion's Club pancake breakfast now.
Oof, thanks for sharing and the well wishes.
I'm funding this myself, and my current approach (hopefully!) avoids most of the red tape. I'm leasing fiber from a local ISP for the colo <-> my home connection, and once I have myself as a successful "customer" of my own ISP, I'll start doing the last mile build out, which is where I expect the red tape to begin.
But I haven't decided if I'll do fiber or wireless, and if I go wireless, I might be able to avoid pole agreements entirely by just working directly with my neighbors. The problem is that our area is pretty heavily wooded, so I'm not sure if I can place antennas high enough to cover a reasonable swatch of the area.
> If I was more charismatic (and wasn't 1000% sure there were pictures of me doing drugs when I was in college), I would consider a run for state office, because it's a shit show for small towns here.
Loads of politicians have come back from worse! Don't let that hold you back.
I'm making an app for self-tracking. Combining elements from habit trackers, health logging and journaling. Built for rich customization and local-first. Want to be free of rigid structures of many existing apps while providing a better UX / usability than using a spreadhsheet.
Landing page + waitlist: https://dailyselftrack.com/
I'm trying to learn 3D scanning and printing: I have a few small projects that I want to do to develop the skill:
I want to 3D print a shell that goes over my car fob: I keep leaning on it and setting off the alarm. The shell would make sure the buttons never get pushed.
I want to 3D print a sleeve that keeps the NCAS dongle in my car charger. I really wish there was a dongle that stayed attached with screws or similar.
This sounds interesting and practical. Where can we follow your project?
Working on a computer algebra system at http://axcas.net
I released some public domain code for computing Groebner bases (F4 and FGLM). I'm hoping these routines will find their way into more systems.
I’m working a Garmin watch app to query all the rich data on the watch (health, physical, environmental, location sensors) from the watch + general AI assistant. Privacy focused using your own keys and Gemini. API calls direct from watch - no backend. https://untether.watch
I'm working on https://pipestack.dev, a workflow automation platform where you bring your own code - as Wasm Components.
Think n8n, but you bring your own code and optionally even your hardware to execute pipelines.
i'm building a camera app that can connect two iPhone peer-to-peer, no server needed.
https://apps.apple.com/us/app/onesnap/id6754680962
Building vet. The goal is to automate open source package vetting beyond just CVE but actually identify code capabilities, malicious code and other security sensitive attributes through code analysis.
https://github.com/safedep/vet
I'm working on Flavia, an ultra-low latency voice AI data analyst that can join your meetings. You can throw in data(csv's, postgres db's, bigquery, posthog analytics for now) and you just talk and ask questions. Using cerebras(2000 tokens per second) and very low latency sandboxes on the fly, you can get back charts/tables/analysis in under 1 second. (excluding time of the actual SQL query if you are doing bigquery).
She can also join your google meet or teams meetings, share her screen and then everyone in the meeting can ask questions and see live results. Currently being used by product managers and executives for mainly analytics and data science use cases.
We plan to open-source it soon if there is demand. Very fast voice+actions is the future imo
https://www.tryflavia.com/
What kind of plan do you have with Cerebras? It seems like something like that would need one of the $1500/month plans at least if there were more than a handful of customers.
This sounds amazing. A demo video would help me finish sign up - I can’t try it without hooking it up to real data, and I don’t want to for a test.
Great feedback thanks! We have added a synthetic e-commerce dataset as an example when you sign up so you can test it without your data first. Will also add a demo video ASAP.
I've been writing https://urbanismnow.com weekly for a year. The idea is to bring you the best ideas from around the world to inspire action where you (c)are.
It's been going well for a side project and now I'm thinking of expanding to have a directory of urbanists on a map so you can easily find people involved in the local discourse and how to get involved.
I recently have gotten into the "drag and drop" forms of programming like Node-RED and n8n.
Obviously, anyone here who has read my posts knows I know how to write code, but having a bunch of built in connectors that are agnostic to each other with the Oauth and the like being somewhat plug and play allows me to iterate on some ideas a lot quicker.
I installed an n8n instance on my server, and have become kind of addicted to making different Discord bots, and I'm having more fun with this than I thought I would. 95% of the stuff on there is basically drag and drop, and when I need more elaborate logic then I can easily drop into JavaScript. I am looking into writing new nodes for different services, and I keep having new ideas for different stuff I want to build.
Working on https://greatreads.dev/
A place to find great blog articles by regular folks related to dev/tech world.
Wondering about the best way I can add a weekly newsletter built on top of the content currently being ingested and still looking for more sources to add to the database (let me know if you have any good recommendations).
After one of my cyber security research caught some attention in media, am now working on -
Open Source Vacuum Robot firmware
https://github.com/codetiger/VacuumTiger
I’ve been making a story-based podcast for Spanish language acquisition with accompanying activities called https://listenreadinteract.com
There’s a free course for true beginners with no login/sign up required. https://listenreadinteract.com/start
https://www.saintbeluga.org/
I was a YC founder in 2006 and now work as a data scientist full-time, but on the side I also do Christian apologetics, helping fellow engineers/scientists/mathematicians seek answers to life's deepest questions.
Some cool articles for the HN crowd:
- [published several days ago] Medical miracles in Lourdes, France recognized by the Catholic Church 2018-2025: https://www.saintbeluga.org/our-lady-of-lourdes-immaculate-c...
- My interview of Evan O'Dorney, a three-time Putnam Fellow and two-time IMO gold medalist, who converted to Catholic Christianity: https://www.saintbeluga.org/veritas-part-i-conversion-of-a-p...
- In-depth scientific overview of Eucharistic miracles: https://www.saintbeluga.org/eucharistic-miracles-god-under-t...
- Conversion testimony by Harvard astronomy professor Karin Oberg: https://www.saintbeluga.org/veritas-part-ii-conversions-at-h...
If we collect DNA from eucharistic miracles can we clone Jesus?
I know you're being facetious, but DNA tests in the Tixtla case said the DNA was too degraded to analyze.
Building https://github.com/openrundev/openrun, a platform for declarative deployment of web apps.
OpenRun runs as a web server, which does GitOps driven app deployments. You can currently deploy apps on a standalone machine, on top of Docker/Podman. Working on adding support for deploying on top of Kubernetes. On Kubernetes, OpenRun will replace your build jobs (Jenkins/Actions etc), CD (ArgoCD etc) and IDP (Backstage etc). The same declarative config which works on a standalone machine will work on Kubernetes, with no YAML to maintain.
Today I'm hacking on automate-terminal, a command line program and Python library that abstracts the various terminal emulator automations (iTerm2, WezTerm, Kitty, tmux) into a single API. Mostly made for use by other tools. https://github.com/irskep/automate-terminal
I'm working on fighting IBM's patent trolls. IBM slapped the words 'AI Interpretability' on Gauss' 200 year old continued fractions and was awarded a patent.
Now they can charge rent if they encounter a continued fraction library in the wild.
It's bizarre
Interesting.
Do you have a link to the patent?
Here it is: https://patents.justia.com/patent/20230401438
On Google Patents: https://patents.google.com/patent/US20230401438A1/en
The authors simply implement a continued fraction library in Pytorch and call the backward() function on the resulting computation graph.
That is, they chain linear neural network layers and use the reciprocal (not RELU ) as the primary non-linearity.
The authors reinvent the wheel countless times:
1. They rename continued fractions and call them ‘ladders’. 2. They label basic division ‘The 1/z nonlinearity’. 3. Ultimately, they take the well-defined concept of Generalized Continued Fractions and call them CoFrNets and got a patent.
IBM's lawyers can strip out all the buzzword garbage if they feel litigious and sue anyone whose written a continued fraction library. Because, that's what the patent (without all the buzzwords) protects.
Thanks for that. That is patently absurd.
You sent me down a rabbit hole. In trying to track it down for myself I read a couple of others that I thought might be it, and was stunned by how obtuse these patents are.
What sort of leverage does this stuff provide? You mentioned "charge rent". What does that look like?
Honestly, I don't even know where to begin. It's insane IBM owns the patent to continued fractions.
If you wrote a continued fraction class in Pytorch and called backwards (or even differentiated the power series) then you're infringing on their copyright.
Every time I talk acquaintances, friends and family members about finances I'm always shocked at how little people know about basic things like tax brackets, 401Ks, IRAs, ETFs, compounding interest, debt management and etc. So I decided to write a financial literacy/education book with a bit of humor and easily comprehensible language to distill some of these topics. I'm about 1 month into it and try to write a chapter a week.
I'm working on https://mimicmarketer.com It allows you to define different personas that you can then test marketing on. This allows you to see how different personas will interact with your marketing. Currently, it has a feature that allows you to define basic personas and test them against two types of copy, as well as a tool that grades your email subject lines and bodies against a generic persona, assessing the likelihood of user interaction with the content.
My other project is https://eggexplorer.com This is a site I wish I had when building out my flock of chickens. It allows you to see the different characteristics of chickens and which hatcheries sell each different breed. You can also see which hatcheries sell hatching eggs for each breed as well.
Slowly but surely:
- A learning tool in Python for Arrays and Algorithms
- A prototype agent-based configuration management system in Perl
- Trying to reinstall Arch Linux on a laptop the second time around (lost my install notes :D)
Mostly doing all of it for learning purposes.
I'm building an analytics and attribution platform for onchain apps, named Formo.
https://formo.so
Think Google Analytics + Posthog designed for crypto users and apps!
I've learned a lot about data engineering and analytics in the past year.
Created a web game: https://www.teqgame.com
I really liked the concept of games like cards against humanity, quiplash, whose line is it anyway etc. However, there was no virtual way to play it with a group of friends. Quiplash required steam setup (which was not possible on my corporate mac). So i built this as an alternate to build upon the formula.
[still in alpha phase so lots and lots of bugs]
After creating the feature request for Claude Code hooks[1] a few months back, Cupcake is nearly ready for release.
Cupcake is a governance/policy-enforcement layer for agents. Its innovation is binding OPA/rego to agent runtimes (via hooks).
I do not believe we will every strictly rely on "better" models in the wild without deterministic guarantees or ways for enterprises to factor in their own alignment - system prompts dont cut it.
https://github.com/eqtylab/cupcake
Stay tuned for the formal release here in a couple of weeks.
[1] https://github.com/anthropics/claude-code/issues/712
Cupcake GitHub: https://github.com/eqtylab/cupcake
Clever name, and concept.
I'm resurrecting peer-to-peer Matrix (https://arewep2pyet.com) thanks to the Dutch government, who started funding it in October.
The main question is which P2P overlay network to use, if any: the prior incarnation used Pinecone (a variant of Yggdrasil), whereas this time we're pondering keeping it simpler and more scalable and using Matrix itself as the backbone to connect together smallish local P2P meshes - so by default you try to route via Matrix, but failing that you look on your LAN or BLE to see if you can talk directly to whoever you're addressing. Time will tell if this works :)
If you are open to suggestions, libp2p has a good amount of sdks for several languages and you can integrate kademlia for peer discovery
We are open to suggestions :) And the very first generation of P2P Matrix was indeed built on libp2p (and Protocol Labs led Element's Series B). However, the thought experiment here is whether we can get away without a full global P2P overlay at all in the interests of keeping it simple & stupid. We might well end up back at libp2p tho!
Edit: another option on our todo list to look into more is Iroh (https://github.com/n0-computer/iroh)
- Zig bindings for AVIF/HEIF
- Local-first app for comparing hardware builds, down to the individual component feature level: specs, benchmarks, even cpu extension support, lanes, how many speakers in X laptop, dolby atmos? screen panel manufacturer(s), etc. Basically, no-nonsense real product comparison for transparent and fast decisions.
My friends and I have been hacking on http://dateit.com for a while. It's an event planning app (works best on iOS and android, but there is a web app) with lots of fun features:
We started working on this all the way back during the Covid lockdown when we wanted to capture that "facebook events" experience without the facebook.It's grown into something much more than our original idea. Most of the features are free and we have a fair pricing model that doesn't nickel-and-dime you like many of the competing apps do. Would love your feedback!
I am finally making my own blog. I have been only planning for ages. I found that I had a lot to say for the past years working on AI, and I want to record them somewhere. I do not expect a lot of visitors or at all in fact. The blog is going to be just for me to remember stuffs and to keep track of them.
I am using hugo to build suckless static pages. LLM helped me so that I don't need to read all their docs. I haven't finished it yet nor posted a single blog. But there will be one soon.
Currently working on getting back into a fitness routine. I got into this habit of hacking on side projects in my very little spare time but I have realized taking care of my body will pay off far more than any project
All the best - the best project to work on! I recently tested a walking pad with a standing desk at my friends place - I was surprisingly productive being able to walk while using the computer. Will be investing in a standing desk and walking pad for my home office.
I'm working on a boardgame with the help of AI. It's way too easy to create placeholder art with an n8n pipeline, but GPT-5 regularly fails at writing and debugging LaTeX which I'm using for all of the card creation.
Specifically, TikZ is often outside the ability of GPT5 to successfully write or debug.
I'm working on a book about using WebViews for cross-platform music software GUIs. It has a particular focus on performance, which I gave a talk about at the Audio Developer Conference last year:
https://www.arthurcarabott.com/adc-2024/
As part of it I am building a code generator to generate shared type definitions in C++ and TypeScript (plus serialization, comparison and cloning).
Working on Ultrazon.com ecommerce store using AI
https://Ultrazon.com
Working on a free alternative to Masterclass - compiled from YouTube clips!
Currently have two binge-able mini courses on How to Start a Startup (could be relevant to folks in here)
Here it is: https://opencademy.com/
Experimenting with various AI models via GitHub Co-Pilot on an extremely niche project to see how far these models have progressed. Used like ~60% of the premium quota to develop the following projects:
website: https://murajah.pages.dev/
Play Store: https://play.google.com/store/apps/details?id=com.murajah.we...
Manual audio splitting tool for the above project: audio-splitter-6b3.pages.dev/
I've always been skeptical of AI-generated code. This is my first experiment with AI agents, where the full code base, implementation, debugging, and deployment are done using AI Agents MCPs.
Used VS Code all the way, i.e., all the source codes, including the code to generate the Google Play Store APK. I only reviewed the source code before committing and helped debug by suggesting ideas/algorithms.
Mostly used Claude Haiku 4.5 like 75% of the time, where it failed, switched to the sonnet 4.5 or GPT 5 codex. Interestingly, when debugging, sometimes one model struggled even after numerous iterations/feedback loops, but then the problem was solved instantly as soon as I switched to another model.
The source codes are available here:
https://github.com/wasi0013/Murajah
https://github.com/wasi0013/audio-splitter
Initially, I thought the audio splitting could be done automatically using some AI models from Hugging Face or Whisper. But the audio files have some complex repetitions; the output was miserable.
So, for now, this splitting is done manually using a Web UI (The audio splitting tool splits the large audio files into multiple small audio files, think of it like a long paragraph is split into multiple sentences.)
I will attempt again to automate this splitting task using AI, after drafting a game plan for tackling the challenges. I'm thinking of using energy drops and other similar factors to create segments.
Bank Rank — an automated bank account that dynamically allocates your money across the best accounts to maximize your rate at all times.
We’re doing an alpha launch in Q1 2026, and if you’re interested, sign up at bankrank.io/waitlist or email bankrank.alpha@gmail.com
I made an LLM-assisted DnD character sheet tracker! It's up here:
https://www.csheet.net/
And the repo is here:
https://github.com/igor47/csheet
If you play DnD, I would love feedback! Feel free to leave it as GitHub issues or discussion.
If you don't play DnD, you might still find the repo interesting. It's hono on bun, I render jsx server side and client side is all htmx. I use vercel's ai toolkit for the LLM interactions, which are super fun and work really well. I think this is a great use for AI actually. I've structured the code so the same services can be called either by the user via forms and routes, or via LLM tool use, so for every action in the code you can do it via either LLM or "manually".
The LLM usage is fun and interesting. What model are you using, and how much customization are you doing to integrate with the app and maintain character?
I suggest adding an export function to make the characters more portable. Maybe export to PDF as well as JSON.
Amazing! I was planning on making a tool just like this, will take yours for a spin!
Working with a group of friends on a "microcontroller-for-makers" kind of thing called the MakerPort. (https://makerport.fun) Sort of similar to an Arduino or micro:bit, but uses the MicroBlocks programming editor (https://microblocks.fun) created by John Maloney, who was the original team leader for Scratch at MIT for 11 years. The hardware includes an mp3 player, I2C ports, accelerometer and true capacitive touch sensors.
I continue to work on My Financé, my personal finance tool.
I’ve been struggling to find substantive traction, so I’m trying to niche down to make the tool really helpful for people who want to quit their jobs.
I built a rudimentary planning and forecasting engine, and am trying to run paid ads to see if the signals resonate with people. I don’t love ads, but maybe trying to understand them will further inform my opinion on them.
One thing I would love to come up with is a way to make the app fully local first, while continuing the ability to sync accounts via plaid. It would be great to not be able to see people’s data at all. Im trying to figure out if there is a good user experience I could provide while minimizing the amount of data I actually have access too. Maybe this feature won’t matter to my primary customers though, I’m not really sure.
I still have a ton of fun working on it, and if it never really makes any money I consider it a great success for my personal learning.
Link for the curious:
https://myfinancereport.com/
I've working on an AI Thumbnail Generator for making YouTube thumbnails and social media images the last few months. https://thumbnail.ai/
Nice work. I am bookmarking this. Looks really good!
Building a docs website [1] for my speech-to-text CLI tool, hns. I use it 5-10 times daily to transcribe my voice, and a few developer friends I've shared it with have also adopted it for daily use. They like that it runs in the terminal and keeps all data local. So, felt like I should write down guides for new users to get started quickly and to highlight key use cases.
Building this documentation website using Docusaurus. This is my first time using Docusaurus, and it feels like a very nice tool for quickly developing a documentation website.
[1]: https://hns-cli.dev/
World model for AI agents. Doing process mining of missions, operations and logistics to transform them into digital twins. AI agents can then leverage these digital twins as world models for control or prediction.
I kinda gave up on building apps for Rocknix (since there is no easy way to distribute software) and instead have been looking at my apple watch. I ported over some software for workouts that is streamlined. I'm working on a way with MDNS to sync data to my Linux PC automatically when I'm at home.
If it works out, maybe this could be a way for me to replace the compromised Apple Music app with something that actually syncs to my music on my desktop.
I've just released my beta of FURS.
FURS does for Forth, what headers do for C, namely provide all the embedded configuration information inside a Cortex-M MCU, for the up to 100 inbuilt peripherals.
Without this data, neither C nor Forth (or any other language) have any clue about how to use the peripherals.
FURS does this by intercepting the Forth user source as it's uploaded to the on-chip compiler and transforming it into language the MCU inherently understands.
The Forth user source code is not altered in any way.
I've used the Fossil DCVS for the entire FURS project so that all the flowcharts, pictures, code, user doc, trouble-ticket, wiki ... everything is contained in the ONE FILE, under 5MB.
This one file gives you a web server so all you need is a browser to easily view all the above from the main menu.
Howto: https://sourceforge.net/projects/mecrisp-stellaris-folkdoc/f...
I am working on PocketWise (https://pocketwise.app) a lightweight personal finance tracking app. Goal is to make double entry accounting simple and approachable for everyday use. It’s my first project of this kind, so I’d really appreciate any feedback.
I’m building https://unrav.io : A tool to fight information overload.
It lets you turn any article, YouTube video, or PDF into summaries, mindmaps, podcasts, chat conversations or infographics that match how you learn with just one click.
We just launched this week the Chrome extension so you can do all this in one click on any page, no login needed (with generous freemium usage).
https://unrav.io https://chromewebstore.google.com/detail/unravio/mbnapibcjcf...
Would love feedback from fellow builders.
I'm making a game finally! Merge-three + village sim.
Hoping to actually take this one to something polished as opposed to the many half-finished prototypes littering my git repo over the years. I've discovered (always knew?) that heavily cutting scope is the best way, and been successful thus far.
It gets pretty boring/unbalanced by ~150 turns, but I have some ideas on how to fix. I'm still playing with ways to help de-clutter the board and make use of the economy aspects.
After some false starts with ai-gen art, I had fun learning to color the pixels myself. The process wasn't as scary as I'd thought and the results are better than I hoped.
https://tower3.dreamofninjas.com/
Inspired by TripleTown from the wonderful studio Spry Fox.
I'm building a fun writing app that mimics the feel of a real typewriter.
Free on iOS + iPad + macOS (Catalyst), and I'm working on adding additional skins, premium features, and Android soon! :)
https://retrotype.ink/
I'm working on a (hopefully) better version of Spelltable to play Magic: The Gathering with my friends: https://cardcast.gg.
I think I got all of the important bits in place, now just working on improving the quality of life experience and bug hunting.
A kernel extension-less sshfs for macOS. I tried using FSKit and got halfway before I felt too constrained by the extension security model (must be app sandboxed, must be approved by the user in system settings). Now it’s just a standalone command line binary that doesn’t require any special permissions since it proxies NFS to SFTP. Everything “just works” and performance is reasonable
Interesting... the way I workaround'd non-macos-native-sshfs is by using docker and mounting a local folder, do you plan on publishing your tool?
Check out my project at https://www.MobiusClock.com: A 3D WebGL Clock on a Möbius Strip that shows 24hr time on a 12hr face. The hour indicator follows the edge of the strip, thus must make 2 turns to return to its starting point, giving you a 24 hour clock. The minute and second indicators move along the middle of the strip and thus return to their starting points in only one turn. Has the ability to rotate!
Just added a new feature: a 'Fast Mode' button to temporarily speed up the hands, which helps visualize how the slow-moving parts work, how the hour indicator moves along the edge. Would love feedback on the implementation.
I'm thinking a lot about the ARC-AGI ML benchmarks, especially the "shape" of the dataset and what that says about how it should be solved. I think there's good reasons to believe that deep learning - at least differentiable SGD backprop style - is a bad fit for this specific benchmark, due to the tasks being almost entirely discrete symmetries, and also having so little data to approximate the discrete symmetries with continuous ones (considering deep learning to be the learning of continuous symmetries). I think that a more explicit and discrete approach is the way to go, and it's possible to build something surprisingly general and not heuristic-based even without gradient descent, guided by minimum description length to search for both grid representations and solver functions. I'm looking for teammates for ARC-3 so hit me up if this sounds interesting, I'd love to chat!
I made a viewer on my website to build intuition for my preferred perception algorithm which is entropy filtering + correlation. Pretty neat to check out the heatmaps for random tasks, there is a lot of information inherent in the heatmap about the structure of the task: https://synapsomorphy.com/arc/
Built a web interface for Magic: the Gathering draft pick advice, trained on top player data.
Website: https://statisticaldrafting.com
Turns out to be a small niche, but I enjoy it!
Porting my 2014 WebGL game (https://phoboslab.org/xibalba/) to the N64: https://x.com/phoboslab/status/1982883072292069641
How are you doing this? Do you have a blog going over the work or a Github page? Just curious, it's an interesting project.
I've been working on a sillier project lately. Green teeth!
Lumina has made a probiotic strain that is able to, theoretically, prevent cavities. I don't care that much about, but I do think it is a neat strain that can likely colonize your mouth. I'm genetically engineering it to express sfGFP, which would theoretically make my teeth fluorescent green under black light. Would be fun at raves! Also, if I make out with anyone, you could theoretically see changes in microbiome composition just from green-ness. I do wonder how much microbiomes are shared while kissing: this would be an example of a way to directly measure that, instead of just measuring on proxy like much microbiome research
That sounds so cool. Could you tell me more?
I'm working on boiling the ocean - we're building a new CRM to compete with some of the big players. I tried very hard to avoid doing this, but I've helped enough business owner friends set up CRMs to realize there's MUCH to be desired. My goal is to create a CRM that people rave about - something that is very rare. Pretty much everyone I help views CRMs as a necessary evil. Our bold challenge is - can we make a CRM that is delightful to use?
Of course we have to slap "AI" on it in this market, but we plan on adding AI features that are actually thoughtful and not just a glorified chatbot.
i'm currently unemployed but i have experience as a CSM and account manager for multiple companies (all have used Salesforce) if you ever want feedback or bounce ideas.
This would be awesome, we're in hardcore MVP build mode right now - when we get closer to launch I'll hit you up.
I'm a little delusional, but I think there's ground to steal back from Salesforce. Most folks I talk to hate how complicated Salesforce is (one even calls it Salesfarce). I've heard a story or two about smaller companies trying to adopt it and wasting a hundred thousand or two implementing Salesforce only to have it never get used. On top of that, you need to train your employees to use Salesforce effectively.
The key is simplicity, building a CRM that anyone can instantly understand just by looking at it. This is insanely hard but I think we'll pull it off. I'll show you more what I mean when I reach out. Thanks!
Sounds interesting! As someone who works in Revenue Operations the CRM space is ripe for disruption, especially around using CRM data to help sales teams explore the data.
I vibecoded a POC of what something I think would work (around the latter part around exploring the data). Need to complete it and start testing it.
Thanks! I totally agree. It starts with getting the basics right. I think this is where most CRMs get it wrong - getting users to correctly enter and maintain clean data is a challenge. We're trying to build smarter schemas for contacts, quotes, sales, etc. but we're also making a really intuitive UI for easily updating information. Further still, we're trying to automatically fill in data for the customer. Many fields can be automatically inferred based on the context of a deal.
Once we get this right, I think the next step is exactly what you said, building really good tools to explore the data. Making it easy for non-technical users to run machine learning on their data to make business decisions or see cool visualizations. We realize that so many of our customers want to know this stuff and have no way of getting at it! We've got a lot of ideas for both visualizing and analyzing the data. I think there's a ton of potential for cool things here. Heatmaps, spiderwebs, interactive charts, etc. Stuff that brings the data to life. One of the common asks from some of our early customers is a heatmap of the world to visualize their sales/reach and see changes over time. Visualizing progress per sales region, etc. I think sales especially has a lot of opportunities for better lead generation and qualification as well.
WebGPU Snippets - https://webgpusnippets.vercel.app/
https://github.com/hsnice16/webgpu-snippets
DIY grid-tied residential solar+inverter+battery. Trying to design the solar arrays' tilt mechanism now for lifting/lowering 5 panels at a time in winter (60-degree winter angle, 35-degree spring/summer/fall; ~24" difference). Thinking either two linear actuators, or a single hydraulic jack connected to multiple support beams. The weight isn't much, but I want a way to lift entire top edge at once to prevent twisting. Linear actuators are slightly more money and easier to build, but require power and weather-proofing. Jack is cheaper, but more complex to distribute force. Wondering if there's other options. (winch would require more robust/taller rear posts, seems more complex, might shade rear array)
Tilting them vertical or nearly so is very useful if there could be any hail, that might be a good idea to support.
What about compressed air? It might not be too hard to find a small brushless low power air pump that could drive pistons directly.
You could mount the pump controller onto the back of the panels and use an accelerometer to measure angle, and run the pump until it's where you want it.
You'd probably need to do some testing and make sure it couldn't get jammed, then build up pressure, then suddenly unstick and move unsafely.
"Projects With Everyday Dave" has done quite a few tests, and may be worth a look:
https://www.youtube.com/watch?v=I-Fz5T5c0OQ
Best of luck =3
Thanks! And I've seen Dave's tests. Since I don't need a fixed angle, do need to maximize production (due to a limited sun horizon), and only have 10 panels, the best option is adjusting angle during winter. The N-S orientation is a really poor performer. (Notice that his test is 30-degree tilt vs N-S; 60-degree tilt will provide 20-25% more power than 30-degree at my latitude, without even considering bifacials w/snow. The only thing that would produce more is an active tracker, but I've got other things to do, and a December deadline...)
Frustrated by the complexity and high overhead of most monitoring tools, I wrote Simon.
It’s a single binary, dependency-free monitor in Rust that does it all: metrics, Docker, alerts, and file browsing. While maintaining a minimal footprint for embedded systems and other constrained hardware.
https://github.com/alibahmanyar/simon
I just launched a 10-Bit Video Thumbnail Provider for Windows.
Windows does not natively support rendering thumbnails for 10-bit videos, which are commonly produced by cameras like the Sony A7IV.
When I started working on a short film the video clips were piling up on my hard drive. Opening them one by one to find what I was looking for was tedious.
I could not find a reputable solution to this problem, so I started a company and built one. I went through the process of EV Certification to have the installer and executable code signed.
I hope to be in the Microsoft Store soon.
I'm also building other utilities with similar purpose.
https://ruptureware.com/thumbprovider
Originally started in 2012, I’m (still) building log.soccer - a stat-tracking tool for amateur soccer players. This is the third or fourth iteration of the site, which until this year mostly served as a glorified résumé project to showcase the latest framework or tool I had just learnt.
Thanks to ChatGPT, my productivity went through the roof this year, and I finally shipped an MVP that might actually be useful.
It’s a standard Django + React + AWS stack. My raison d’être is to build an Apple Watch app that tracks match scores in real time. 2026 hopefully.
https://log.soccer
I’ve been working on an online LPC speech encoder as part of my embedded sound tools: https://buzzer-studio.atomic14.com/
It’s starting to sound pretty good. I’ve had quite a lot of help from the community with lots of useful feedback and suggestions. It’s been fun.
The Daily Baffle, a site with all sorts of daily puzzles including one clued daily by NYT-published constructors.
Can check it out at https://dailybaffle.com
I'm still working on growing the audience. App coming soon!
This is awesome, shared with a few people I know would like this.
We are building end-to-end accessibility compliance tool[1] that will take care of auditing, remediation, verification and generation of ACR/VPAT.
Because of the well bound nature of the problem space, we are able to unlock a lot of power from LLMs and put together a good end-to-end product that delivers the promise.
Still early days. I know there are lot of folks who care about a11y. I would love to chat and learn from your experience.
[1] https://workback.ai/
Working on therapy software, for porn addiction.
https://zenstreak.app/
I've been working web scraping using LLMs, I just shared one of the libraries I created to get structured data from arbitrary pages: https://news.ycombinator.com/item?id=45870231
Instead of sending the page's HTML to an LLM, Hikugen asks it to generate python code to fetch the data and enforces the generated data conforms to a Pydantic schema defined by the user. I'm using this to power yomu (https://github.com/goncharom/yomu), a personal email newsletter built from arbitrary websites.
I'm working on SuperCurate (https://getsupercurate.com), which is geared towards note retrieval and curation rather than note creation. Think filing cabinet for your notes, web clippings, images and PDFs.
I wanted fast search and filters for my Evernote archive so I could drill down and surface exactly what I was looking for.
There's also a Web Clipper extension for Chrome.
Demos:
Search and curation: https://www.youtube.com/watch?v=z4QSIoUL4Uk
Web Clipper: https://www.youtube.com/watch?v=8F7QoC7X3fs
Search inside PDFs (jumps to page + highlights snippet): https://www.youtube.com/watch?v=t0X9sD-938Q
It's free while in beta, would love feedback if you try it.
Coding agents that you are building should get the link to this post and build everything mentioned.
My friend and I working on PennyPost, financial analyst that solely focuses on your spending.
It looks at your spending across all your accounts, categorizes, identifies patterns, trends, runs predictions and sends weekly/monthly summary email.
No apps or dashboards, just insight.
Plug in once and forget. Takes 5 minutes to keep track of your spending.
We are still at early stages but you can check it out here
https://pennypost-landing.vercel.app/
Appreciate any feedback you might have!
I’m working on a performance capture library for Python because I often need to know the performance of backend systems I maintain. I frequently build tooling to capture performance and save it for later analysis. I/O operations get costly when writing lots of data to disk and creating good real-time analytics tools takes a lot of my time. I wanted a library that captures real-time performance analytics from Python backends.
https://github.com/jakeroggenbuck/kronicler
This is why I wrote kronicler to record performance metrics while being fast and simple to implement. I built my own columnar database in Rust to capture and analyze these logs.
To capture logs, `import kronicler` and add `@kronicler.capture` as a decorator to functions in Python. It will then start saving performance metrics to the custom database on disk. You can also use the middleware for FastAPI.
You can then view these performance metrics by adding a route to your server called `/logs` where you return `DB.logs()`. You can paste your hosted URL into the settings of usekronicler.com (the online dashboard) and view your data with a couple charts. View the readme or the website for more details for how to do this.
I'm still working on features like concurrency and other overall improvements. I've added a lot since the last time I shared on HN. I would love some feedback to help shape this product into something useful for you all.
Thanks! - Jake
I’m building Culink — a platform for curated link collections.
Not random bookmarks, but organized and shareable collections you can actually discover. Think Pinterest for links.
Day 13: 27 collections live, and people are already creating their own — which is exciting to see!
https://www.culink.io/discover
We've been tinkering with building realtime talking head models (avatar models, etc.) for a while now, and finally have something that works (well enough)! Operates at ~2x realtime on a 4090, significantly faster than that on enterprise grade GPUs.
You can try it yourself at https://playground.keyframelabs.com/playground/persona-1 and there's a (semi)technical blog post at https://www.keyframelabs.com/blog/persona-1
The main use case we designed for was language learning, particularly having a conversational partner -- generally we've found that adding a face to the voice really helps trigger the fight or flight response, which we've found to be the hardest part of speaking a new language with confidence.
But in building out the system around the model to enable that use case (tool use on a canvas for speaking prompts and images, memory to make conversations less stale, etc.), we think there's potential for other use cases too.
Tools to help my mental health tracking[0], and sharing with others how I manage my limited amount of energy[1]. They're kind of related, since mental health impacts my energy, so I've needed to prioritize and really make sure I'm spending my time and energy on things that matter. Usually, there's a good mix of things I enjoy doing with things I gain a lot out of. I've spent a lot of time thinking about this!
I've used my app in various forms for around 5 years, rewritten multiple times. But now I'm creating surrounding tooling to help others put my mental model for personal life prioritize to use. I'm writing in the "Saving Spoons" Substack as I go, trying to explain why and how I do things, with advice for others trying to do the same thing.
[0] - https://github.com/eeue56/gobaith/
[1] - http://savingspoons.substack.com/
Meta question:
Is there a cadence for these threads? I had in mind to "be prepared" to post in November's with what I'm working on, but I expected it to come around on the 15th (mid-month).
What I'm working on:
- skuilder - (skill builder) - https://github.com/patched-network/vue-skuilder - an AGPL framework / toolkit for SRS++ based interactive tutoring systems
- https://letterspractice.com - a low cost, hopefully high quality early literacy acquisition app, targeting ages 3-5.
- https://flutor.app - an app to learn the flute
(The proprietary apps are built with the toolkit).
I've struggled to pitch or articulate the vision here, but my latest pithy attempt is: scaling self-actualization by mechanizing the nested loops described by Anders Ericson's 'deliberate practice' - Inner loop: individual learners maximize their skill uptake velocity and performance peak by adhering to domain specific best practices - Outer loop: domain specific best practices get refined according to innovation or serendipitous discoveries from the inner loop (eg, someone is observed to beat out prior best practices)
As mentioned, I'm flat-foot posting here, so the pages aren't all prepped. https://flutor.app/dbg and https://letterspractice.com/dbg show some of the innards. Not linked, but I'm especially fond of https://letterspractice.com/dbg/juggling - the premise here that as child practices the letters, the letters exemplify the principles of effective practice in alliterative skill domains (juggling Js, batting Bs, flossing Fs (it's hard ok?))
Disenhackifying one of the last pieces of my KaithemAutomation server that still feels not best practicesful.
Device driver plugins used to have a very simple flat key value, strings only format, with a set_config_properties function to tell the host what kind of UI to show.
That's all getting replaced with JSON schemas, with some auto-upgrade shims so old config keeps working.
It's one of many things that now seems completely insane, but made sense when I had way less experience a long time ago!
Also still on and off working on my BLE/WiFi based Meshtastic-alike.
Local only meeting transcription and summarisation for MacOS.
https://localscribe.app/
Loads of similar products out there, but non that did all of: open source code with attested releases, recorded mic and system audio to work with any meeting app and used Apple Intelligence for private summarisation. In beta, and also just released a experimental version with self hosted Ollama support.
Currently working on training language models steered towards certain "states of consciousness".
I have a model trained on publics datasets tied to brainwaves and/eye tracking and text comprehension (have this working well enough to experiment). Now I am training an adapter for various llm architectures to generate text steered to certain neural oscillation patterns (let's call them "states of consciousness" for brevity). I also have a 'rephraser' that rephrases text to elicit these certain states of consciousness. Overall experimenting with creating an suite of tools off my findings with how text relates to the eigenmodes of consciousness. My theory is once I do this I'll be able to do some...interesting things with "AI" agents. lmk if you want to talk about it if you're someone with knowledge in neuroscience/ML. My background is as a Software/ML Engineer so I could use additional thoughts. I do wish I could send a Github/docs which I will soon but this is currently a private project seeking investment for various research/public/private sector applications.
Curated LinkedIn topics + AI drafting: validating before building
I'm exploring building a weekly curation service for professionals who want to write on LinkedIn but struggle with "what's worth writing about."
The thesis: In the AI era, execution (writing) is commoditized. The real bottleneck is editorial judgment... knowing what topics matter before they're obvious.
The concept: Weekly email with 5-7 curated topics (tech trends, policy shifts, market movements). Each topic comes with sources, multiple angles, and context Choose your perspective, AI drafts a polished article
Why I think this could work: I've been manually doing this for myself for years. Pattern recognition at scale is hard to automate, but pairing human curation with AI execution might work.
Target market: ~30M professionals who should be building thought leadership but don't have time to spend on research.
Current status: Validating demand before building. The hard part isn't the AI, it's systematizing the trend-spotting and curation process without losing signal quality.
Bread and butter stuff. Pulling together all of the assorted algorithms and data structures I implemented in C over the years out of necessity - lists, trees, stacks, queues, hash tables, memory pools, etc. - aligning the APIs, cleaning up and merging into a library. It's a background project but super fun. This and several parsers - JSON, some config file formats, and parsers for some GPS / GNSS receiver data protocols. FSMs also always feel like nice, clean fun. And prematurely optimising every bit.
Updating partijgedrag, which is a voting compass based on how parties vote on motions in the Netherlands!
Apart from that I have a personal SaaS idea I want to release soon. Its something that started as a joke but the joke is still not finished
https://unskip.me
Currently working on a website that lets you add any form of troll CAPTCHA to your website or allows you to create a redirection with shortened link.
It is a fun project and oh boy I really enjoy solving CAPTCHAs on random websites.
I don't get it, how do I shorten a link?
I made a direnv-like utility which hooks into your shell to allow using Nix flakes for dev envs / dev shells anywhere:
https://github.com/dfrankland/envoluntary
This helped me bridge the gap between installing packages declaratively via NixOS / home-manager and defining them for each project being worked in via flake.nix / direnv / nix-direnv; which was needed since most projects don't use Nix.
Stopping Agents [1] - LLM agents that stop conversations early and save time.
Because they're trained using imitation learning instead of RL, they're scalable and easy to deploy with your own data (also open-source!).
Mainly targeted at and tested on quickly disqualifying prospects in sales calls, but can be applied more broadly.
[1] https://stoppingagents.com
I’m building Sink It for Reddit (https://gosinkit.com), a browser extension to make Reddit usable on the web. It’s similar to RES with broader support for all the different Reddit UIs (there are 4).
It’s mostly free with only old Reddit features gated behind a one time $5 fee. The app has a few hundred thousand users on the Apple platforms but recently it was invited to join Mozilla’s Recommended Extensions program so I’m hoping to grow the non-Apple user base.
A multiplayer game prototype, written in Rust, using Bevy + egui for graphics. Think of it as a bare bones implementation of a game like Runescape, mostly to test out current LLM capability.
I haven't found much value in LLMs for coding beyond very self contained tasks, but some people speak highly of it, and I want to be sure that I'm not missing out. So from time to time I give new tools a try. This time is "Claude Code on the web".
I've put in an estimated 50 hours so far. It has a client and an authoritative server. The client displays 3D graphics with some placeholder models. From the client, you can click on tiles and move to them, or click on enemies to pathfind and attack them. You can right-click on tiles or monsters to open a menu with options (attack, trade, move). There are some unit tests and a few integration tests.
Right now the issues that Claude has been unable to resolve after a few attempts are: * Attack animations. I'm trying to get it to raise and then lower a rectangular block to simulate a sword attack. It really doesn't get it, and it's harder to write tests for compared to movement and server-client networking. * "Entity interpolation". Rather walking entities instantly moving from tile to tile, movement should flow smoothly.
I have Claude Pro ($20/mo) which let me make a few commits per day. After a few days of that, Anthropic offered $250 in credits to promote "Claude Code on the web". The credits expire after two weeks. I'm now five days into that period and have gone through $50 in credits. It is heavily rate limited and frequently locks me out for multiple hours after only a few interactions, but it's free credits so I can't really complain.
I've been working on two game development projects for the past couple of years.
One project is for building rhythm games in multiple game engines and multiple platforms. Currently, it works in Unity, Unreal, Godot, SDL (or any C++ game engine), and MonoGame (or any C# game engine), and runs on Windows, macOS, and Linux. I'm working on adding Love2d (or any Lua game engine) and Bevy (or any Rust game engine). I have a few local prototypes of it working in Unity and Godot, but nothing public yet. Still trying to figure out what kind of game I want to make with it.
https://github.com/neogeek/rhythm-game-utilities
The other is a general purpose game engine in C++ with SDL. It's far enough along that I'm building games in it, but it's more of an exploration into how games are made than a replacement for Unity or Godot. I suppose it could be eventually, but I'm trying to be realistic with what it can do. One thing I'm pretty happy with regarding this engine is that one of the demo repos will automatically build to WebGL and publish to itch.io when changes are pushed.
https://github.com/HandcrankEngine/HandcrankEngine
I'm working on _prompt injection_, the problem where LLMs can't reliably distinguish between the user's instructions and untrusted content like web search results.
Just published a blog post a few minutes ago: https://alexcbecker.net/blog/prompt-injection-benchmark.html
Good post. Thanks for sharing. I enjoyed it as much as I enjoyed your anime list. I agree on many.
Unwound a couple of things from the stack!
Finished: the 100%-vibe-coded "GPT-5 reviews all my PRs on max reasoning" GitHub app (which is shockingly effective, https://github.com/Smaug123/robocop - probably nothing new for people who already use some product like this, but I like owning my own infrastructure as far as possible, and GPT-5 and perhaps Gemini are the only models smart enough to do this so I can't take this any further).
Currently: back on "write an immediate-mode TUI framework that uses a vdom as its fundamental abstraction" (https://github.com/Smaug123/WoofWare.Zoomies), in the hope that this is the first UI framework that I don't absolutely loathe.
Next: using the TUI framework, write a debugger to inspect the internal state of my deterministic .NET runtime (https://github.com/Smaug123/WoofWare.PawPrint) and to step forward and backward in time.
Next: get the deterministic .NET runtime to a point where a property-based testing framework can identify the deadlock in some very simple buggy multithreaded code. (The framework is not yet able to run Hello World - did you know that's an incredibly complicated program in .NET? - but it can solve a few Advent of Code problems right now, can perform some limited exception handling, limited virtual method dispatch, limited casting between types. Even getting to Hello World might take a year if I'm unlucky.)
I'm working on basi, an alternative/syntax to writing Playwright: https://github.com/zikani03/basi
Trying to keep it simple but I can already feel some "design pressure" to think about making the DSL more complete (language) by adding features like loops and variables. Still early days!
Made a website to host a blog! Right now it's empty except for one post describing the process of setting up the blog. I plan to add more stuff once I finish this semester in college.
Website: https://ngonella.com/
I have a bunch of ideas and small projects I would like to write about, so I'm really excited about this.
Next release of my app (8 years since initial release) - adding minor features and minor bugfixes.
Video Hub App - browse your local video files with a beautiful interface (and scrub-able thumbnails to see multiple screenshots)
$5 for anyone https://videohubapp.com Free for anyone https://github.com/whyboris/Video-Hub-App
I'm still working on WithAudio (https://desktop.with.audio). A one time payment Text To Speech Desktop App. Because I think everything doesn't have to be a subscription.
In October I finished the PDF parser. It was a big challenge extracting PDF contect with correct paragraph breaks on user's computer locally. I'm gonna write about this soon.
Now I'm working on a web extension that talks to the app that run locally on your system so you can use WithAudio in your browser with very good performance, 100% local and private.
Still hacking on some data tools:
DuckDB for stream processing:
https://github.com/turbolytics/sql-flow
Lightweight kafka stream processing using DuckDB as the execution engine. 300MiB runtime can easily handle thousands of messages / second.
Working on a Kafka Connect alternative:
https://github.com/turbolytics/librarian
Right now mongo replication (through changestreams) is supported to kafka. Working on Postgres support right now.
I'm working on Argon Chess, a deterministic chess variant with some degree of cheat resistance (hard to describe to chess engines like Fairy Stockfish) and tons of variety. A week ago, I added a way to play friends online a week ago (a Discord Activity) and a simple Play a Dumb AI feature on its website. You can also print the cards for free for offline play. https://argonchess.com/
Developing a fingerprinting method for identifying music masterings! Like Shazam but to tell what version of an album you have.
The idea being able to compare measurements to see what mastering you're really getting - because they are NOT all equal. With the remasters and stealth replacements on streaming, it seems like every other month I wake up one day and my favorite music sounds worse (or is gone...). Now I can measure it and help find what versions I really want to collect!
I may end up trying to make a fingerprint database/tool that sits in between MusicBrainz and Discogs. That way hopefully the community can standardize and quantify some of this info that only lives ad hoc in Steve Hoffman forum threads or partially on sites like https://dr.loudness-war.info
I’m still working on https://opus.cafe/ outside my full time job.
It’s been a fun 3 year project. Just launched on iOS and am in user acquisition phase. Totally new learnings here! Getting users is definitely the hard part... I can build something all day
A new ping pong paddle design.
My backhand is OK but my forehand sucks. Grip styles for standard handles usually end up favoring one side or the other. I'm making a handle shape that's easier to get the blade angle right on both sides. Hopefully a couple more iterations on the 3D printer and then I can have a functional prototype made.
I'm working on https://yap.town - an SRS based language learning app.
I would say it combines the best parts of Duolingo and Anki. Anki is great for memorizing words, but you don't see the words in the context of novel sentences. Duolingo is great for exposure to new sentences, but it's oriented around "lessons" and SRS is an afterthought. (Duolingo is also not designed for people serious about learning a language IMO, it's too easy and goes too slowly.)
Had to do quite a bit to get it to work well.
1. At first you would think that if you know all the words in a sentence, that should be enough to understand the sentence. But it doesn't work like that. For starters, words can have multiple meanings. The french word "bois" can mean "(you) drink" or "wood". You want to learn these separately. I trained an NLP model (a gemma3 finetune) that I use to understand the manner each word is used in each sentence: https://huggingface.co/collections/anchpop/lexide-nlp-models
2. Even then, what about a sentence like "you'd better not"? Even if you know the words "you" "had" "better" and "not", you still won't really get this. So I use the wiktionary "multiword terms" category for each language to get a huge list of terms like "'d better" , "you better believe it", etc, and teach these in addition to individual words. And then I only show sentences where you know all the individual words as well as all the terms.
This is awesome sir, I'm pretty sure this will get you very rich soon or later.
By the way, I have a suggestion, the examples on the answers could be listenable to keep the brain on a learning mode all the time even on side words
Great idea, done!
And I'm not planning to get rich off of it haha. Right now there's no monetization at all. If lots of people use it to learn a language and avoid wasting their time on duolingo, I'll be happy
I am learning Godot engine, going through the list of 20 games in order to build up my experience https://20_games_challenge.gitlab.io/
I am almost done with flappy bird (2nd challenge)
Why? I love the old arcade and game boy games, and I want to recreate them to my liking. I also love mechanical systems and space rovers, and I want ro build worlds to explore and simulate these things
Thanks for the link to that 20 games challenge. Godot is fun and something I want to get back to eventually!
For the past few months, I've been building a health tracking app called LogBuddy. I got tired of using separate apps for nutrition, workouts, weight, and period tracking. So I built LogBuddy to handle all of it with a dead simple interface. That way, all the heath data I'm tracking would be in one place. Right now there's an Android APK available. If there's interest, I'll publish to the Play Store and build an iOS version too.
Would love to hear your feedback if you try it out!
Here's the github repository: https://github.com/aabiji/logbuddy
Working on developing a suite of apps around photography, from cameras to editors and utilities.
https://heliographe.studio
The goal is lightweight, composable tools with clean interfaces that respect user agency and privacy, provide technical clarity, and make you a better photographer by encouraging mastery over your tools and offering new ways to approach picture making. Also broadly honoring the (almost) 2 century old history of the craft and drawing inspiration from pre-digital processes and approaches.
Got a number of updates to existing apps and new ones in the works, I’m excited for the full long term vision I have that I plan to sum up in an essay at some point.
Currently Apple platforms only but the plan is also to break out of that down the line.
Still working on my digital nomad event and workation aggregator.
But now with travel and visa guides to help remote workers become productive in Japan and South Korea ASAP and give them visa guidance if they want to stay a bit longer.
https://reorient.guide/
https://github.com/FarisZR/knocker
Knocker, an http knock based access service for your homelab that works at a reverse proxy or firewall level.
It's a more convenient albeit less secure alternative to VPNs like tailscale. It's more convenient because it whitelists the enite network, and it's less secure for that reason.
https://recipin.com Recipe extraction and archiving to avoid link rot and blog spam. No tracking, no JavaScript, no AI[0], and just a dusting of CSS. Source available to run your own server if you’d like (https://github.com/bradly/recipin).
I’d like to add importing from a Pinterest account and continue adding support for all the creative implementations of the schema.org recipe format that different sites use.
[0] My partner has a bunch of handwritten family recipes, so I’m trying out an optional extract from a photo of a hand written or magazine recipe that uses AI. Not required and I may pull it out into its own service that spits out schema.org recipes. We’ll see.
https://yournextstore.com https://github.com/yournextstore/yournextstore
I'm building Your Next Store (YNS); it's a Shopify alternative built with React and Next.js.
We provide an opinionated boilerplate tailored for tools like Claude or Codex, so designers and developers can build storefronts faster and more easily. It enforces a clear structure to start from while keeping full control over design, animations, and the overall storefront experience. It’s built on top of Stripe, with our higher-level commerce abstractions, like "add to cart", "checkout", "pay", "browse products" etc; plus a Commerce CMS so merchants can manage everything smoothly once their store is live.
If youre planning to sell something online and want a modern solution, hit me up! :)
a website blocker that uses AI to help you against procrasticantion, you can try it for free: https://tasksentry.app
Correct and performant way to calculate historical value of a portfolio. I want a pure function, but taking a date as input is insufficient because users can edit holdings, and securities can split.
Weighing the tradeoffs of doing this calculation server or client side. That'll be an architecture shift away from my current set of background jobs fetching state and towards something more functional and on-demand.
https://jch.app
https://linog.ph
LINOG.ph is a live earthquake tracker for the Philippines.
The Philippines deals with thousands of earthquakes a year. Whenever the government volcanology and seismology department detects earthquakes, they post it on their official website.
When a major earthquake happens, a huge number of people try to visit the site, causing downtime for up to an hour.
LINOG.ph caches earthquake data from the official government website and the U.S. Geological Survey site, and makes them highly available to the public.
I built this after seeing friends and family donating and providing support for affected families after a major earthquake in Cebu. This was my way of helping out.
Two super typhoons have hit the Philippines in the past two weeks, so I'm also considering adding in typhoon tracking.
Why doesn’t the government put Cloudflare or something in front of their site?
I'm working on my own code review app powered by local or self hosted LLMs. It started as a way to lint my own code and took off from there. It's basically like greptile or co-pilot, but has some things that they don't:
https://drep-ai.org
I’m working on Reflect [0], it’s a privacy-focused app for self-tracking and self-discovery. You can track metrics, run self-experiments, set goals, view correlations, visualize your data, etc.
[0] https://apps.apple.com/us/app/reflect-track-anything/id64638...
Imo, you really should use a pure white background for the App Store screenshots instead of the current greyish background which looks kind of depressing.
I built https://forvard.org/ with Tauri + Svelte; Forvard is a accomplishment tracker where your data lives locally (think of Obsidian but for career tracking), it has a bit of smartness to summarize your accomplishments (without sending your data over the network). I'm working on fixing bugs and adding couple of features!
Frontend framework in JavaScript that requires no build step, relies on DOM and SSR and can be used to build both SPA and hybrid apps without VDOM, js templates, hydration or putting HTML (or worse, css) inside JS code. It'll also have a very sophisticated declarative state manager which makes managing state and ui transitions a breeze. It's basically anti-React.
can you share a repo link please?
Sure: https://code.qount25.dev/qite/qite-js No docs yet, but I suggest you go to test/demo for examples. You can actually see them work if you run it with `node test/server.js`.
State manager isn't there yet, but it's coming.
I'm working on a video / post on how to solve the 1 billion row challenge (https://github.com/gunnarmorling/1brc) and get a competitively fast result while keeping the code readable and maintainable.
So far I'm within spitting distance of the winning entries without using any unsafe code or bit twiddling tricks or custom JVMs or anything like that, and having all the concerns nicely separated and modularized.
Excited to share soon!
My booktube channel : https://www.youtube.com/@NevsBookChannel
Of relevance - my review of Code Complete: https://www.youtube.com/watch?v=QlY0EGWp7rw
https://github.com/codeflash-ai/codeflash/
Codeflash optimizes any Python code for performance by using AI and verification.
We make all human and AI written code super-intelligent by discovering new algorithms and fixing any performance mistakes.
I've been building a little toy computer and assembly language that's interpreted in python. Pretty close to the first release (and introductory blog post) and a lot of fun to build (and learn a bit more about real assembly as I go).
https://github.com/daturkel/dt31
I am building a foundational layer for building C++ apps using Bazel.
I am working on creating a standardized set of paths and third party libraries that work seamlessly across multiple developer teams. Allowing library upgrades to happen transparently in the background. This will enable developers to focus on business specific logic and not have to worry about the intricacies of the build system and allowing to "magically" work in the background. This is allow foray into Bazel and using it as a learning exercise to master it.
Spending all day every day thinking about how to make coding agents better through configs and dev tools
If you're interested in following along, check out https://www.npmjs.com/package/nori-ai
I'm building one project a week for the next 25 weeks for my newsletter. First, I want interesting content for the newsletter. Second, I want to try to grow the newsletter to put out something fun and joyous. The world needs more good fun.
https://randomdailyurls.com
The Board: A feature board that vibe-codes the top voted feature into itself every night: https://theboard.stavros.io
very surprised I haven't seen this done before - cool idea!
Probably because security nightmare
I've been obsessed with reticulum. It's a network over other networks. https://konsumer.js.org/nomadnet-js/ https://github.com/konsumer/rns-lite
The protocol is fairly simple, encrypted by default, and works over lots of interesting transports.
It's really cool, but as far as I know there's no complete C++ implementation for embedded platforms, and I still can't figure out how it actually works.
Does the gossip flooding mean every single node needs to know about every other node in the entire mesh?
I have a project vaguely inspired by this and Meshtastic that tries to make use of existing internet tech, while falling back to local links, instead of trying to replace the Internet completely.
It's very much WIP, I'm planning to get rid of all of the automatic reliable retransmit stuff and replace it with per channel end to end acknowledgment. https://github.com/EternityForest/LazyMesh#
I started working on a Arduino implementation, but it needs more testing.
It works like your address is the hash of your pubkey, and you can announce that or not.
Building https://www.hessra.net/, an authorization system based on the Biscuit token format (decentralized, signed, and attenuable). The goal is to push beyond JWTs and Zanzibar-style policy engines by giving every machine-to-machine request its own embedded, verifiable authorization logic in a small capability token. These tokens can be delegated, restricted, and verified locally with no extra network calls required after getting the token.
Early use case is replacing API keys with identity tokens that expire, delegate, and prove possession and then can be used for easy step up to fine-grained authorization. There's some pretty interesting authorization stuff you can do, like having multiple parties sign off before a token is valid or requiring a series of micro-services sign a token for it to be valid.
I'm working on rebuilding Playwright from the ground up, but focused on automation and self healing using LLMs.
It's called Stagehand (https://github.com/browserbase/stagehand) and we just released v3, which is a total rewrite.
Interesting.
I rewrote Playwright to run completely in a Chrome Extension without CDP or chrome.devtools for no practical reason at all. I started to do it like Forest Gump started running. It can't get past bot protection so pretty worthless from a browser automation point of view. [0]
What I don't understand is why the need to rewrite Playwright instead of just patching it. Playwright (or Puppeteer) has addressed every edge case that has come -- especially race conditions which are a monster to deal with -- up over the years and by the time you do the same you will have Playwright.
Why is rewriting or rebuilding Playwright from the ground up needed?
[0] https://github.com/adam-s/cordyceps/tree/main/pages/side-pan...
Very cool. I make a consulting business out of packaging selenium scripts into windows apps for small businesses, do you have any desire to turn this into a saleable product?
What type of stuff do people pay you to do?
Stagehand is our open source project, but the company behind it is called Browserbase - https://browserbase.com/ where we run headless browser infrastructure as a service. So no interest at this point, Browserbase drives the revenue that funds Stagehand!
I've been working on an open-source platform to build agentic workflows. I really wanted a Figma-like experience for building agents.
Apache-2.0 license: https://github.com/simstudioai/sim
Some weeks ago I launched https://woomarks.com a Pocket replacement.
Here it is hosted on my website https://roberto.fyi/bookmarks/
Currently working on secure file intake for Intercom. Recently spoke to a customer and turns out file intake is just a part of the bigger story, currently thinking about processing files, e-signature, client portals and so much more.
Lesson: speak to customers!
https://www.fibrehq.com/
Nonoverse[1], an iOS puzzle game about nonograms (image logic puzzles).
So far all levels have been handmade pixel art. I’m now testing machine generated puzzles with random “pixels”. This is an interesting challenge because I still want levels to be solvable and fun. I recently released 15 new puzzles like this and I’m preparing a new update with more.
[1]: https://apps.apple.com/app/nonoverse-nonogram-puzzles/id6748...
I'm going all in on my side project CodeBrew, a Java IDE for iPad. Currently working on OpenGL support for 3d graphics, as some schools requested the feature. Also I'm finally pitting some work into aquisition, which has turned out to be much more fun than I anticipated.
Go check it out, its free to try, with a one-time purchase full version:
https://apps.apple.com/app/apple-store/id6475267297?pt=11914...
Working on https://gametje.com (a Jackbox games competitor). Been working on the Android TV app lately. Will probably start creating a new game next week with acronyms similar to the old game Acrophobia from the late 90s/early 2000s.
A new web server written and server management dashboard in JavaScript that is much faster and less complicated than either Apache or NGINX and serves HTTP and WebSockets from the same port.
Porting LevelDB[1] to Seastar[2], for internal metadata storage in Redpanda[3]. Before you ask why can’t something off the shelf be used, seastar has unique constraints around its runtime and its memory allocator that means we can’t reuse an existing library.
1: https://github.com/google/leveldb
2: https://github.com/scylladb/seastar
3: https://github.com/redpanda-data/redpanda/pull/28351
Month three working on my game development resources website: https://raizensoft.com/tutorials/
I write almost daily article about libGDX - my most favorite code-centric game framework. There are now over 100 articles covering topics from basics to advances. I plan to post more because this is more or less a passionate project.
In the future I hope it evolves into a definitive resource for learning game development with Java and libGDX.
Scratching my own (and my employer's, but they don't know that) itch and building a knowledge management system as a nerdy way of spending evenings. I refused to learn JS for years, but turns out it's not as bad as I thought, and TS makes it really nice, plus I like (to my surprise) SolidJS' JSX interpretation quite a lot. Half vibe-coded, half breaking things and learning a lot.
I posted in this monthly thread first time in May when I launched a daily logic puzzle, Clues by Sam. Since then it's grown significantly, and I couldn't be happier!
The game has a farily simple frontend, but there is a fairly complex constraint solving algorithm as part of the puzzle making process. What makes the puzzle quite unique is that you can't "guess". You can only make guesses that are provable by logic. The algorithm ensuring this has worked flawlessly for months now (though I've manually inserted some silly mistakes once or twice).
Today's puzzle is one of the hardest to date. The difficulty resets on Mondays, and then gets harder again towards Sunday.
https://cluesbysam.com
Clues By Sam has become THE morning fixation in our household since I first heard mention of it a few weeks ago - super fun, thank you!
http://www.taxmax.dev - helps companies deduct more engineering spend.
https://github.com/s1liconcow/skyshelve - persistent python dictionary on S3. Used this to create a durable execution layer to do some of the analytics for the above.
I finally started modding a total war game (Warhammer 3). I played the series since the very first title, Shogun and I always wanted to improve the control over units and add my custom AI to not micromanage everything, but assumed it would be too time consuming and distracting from my main work. And well, it likely would have been.
But thanks to LLMs, I finally decided to give it a go and got something basic working in a short time, hurrey for AI assisted coding!
Feels empowering to be honest. No idea if I will really implement the main ideas, that I have since a long time, but I know that I can now if I want to.
A notation and IDE for writing fractal poetry (LambdaConf video https://www.youtube.com/watch?v=adjg2LeQMvk)
A constraint-solver for a novel algorithmic theory of harmony (cadence.is)
a word game based on transformations from one word to another (unreleased)
email my username at thekeyunlocks.us for access to any of them or if you want to talk shop!
https://github.com/benjajaja/mdfried
mdfried, markdown viewer for the terminal that renders headers as Big Text via the new text-sizing-protocol or as images.
I am currently working on a fork of Alt+Tab replacement Switcheroo to show all available windows in a tabular format. Current windows in the center. Apps with many windows on the left sorted by process name (e.g. Excel or PDF windows). Pinned windows such as open emails on the right. https://github.com/coezbek/switcheroo
Vibe coding a programming language: https://github.com/davidkellis/able
I've been building with local AI, on Apple Silicon. It's only 8mb, but runs 30% faster than Ollama.
https://github.com/dinoki-ai/osaurus
I have a comprehensive agent system. Currently working on SIP calls with different models including realtime. https://github.com/runvnc/mindroot
I have two side projects I continue to make some headway on: a practice call platform (practicecallai.com) and an iOS baby tracker / newborn app (https://apps.apple.com/us/app/lil-baby-tracker-newborn-log/i...).
The last month has had its ups and downs.
Ups = some local-area doulas have started sharing the baby app in a big WhatsApp group & growth is starting to pick up.
Downs = my first vibe-coding horror story. For PracticeCallAI, the subscription flow was failing and somehow outside my test coverage, so I've been missing out on new subscribers for the last two months. In an effort fix it, Replit Agent - which I have been loving otherwise - truncated the table that stores all of the user calls. and their database rollback is throwing errors. So that's been fun.
I am working on methods to automate my VC firm. We have a small team and many different tasks to do. I’ve had success with using LLMs to help us automate various projects. But I appreciate any open source tools, techniques, readings, etc. if anyone knows any!
what sort of tasks are you trying to automate?
I'm adding an overly elaborate item and levelling up system to the adventure mode of my chess variant AI sandbox: www.chesscraft.ca
Items have a prefix and suffix system similar to Diablo 2 so I'm having nostalgic fun building it. None of this gives any advantage to the chess games you play. It's just a pointless cycle of gems, items, and experience to get more gems, items and experience. Seems fun so far.
building a location-based game that captures Taiwan's absurd convenience stores density[1]. Players stand at any 7-11/FamilyMart/etc, take a photo showing you can see the next store, walk to it, repeat. Chain as many stores together as you can.
It sounds silly but Taiwan really is this convenient - you often can see 2-3 stores from one spot. Here[2] one route where you can actually link 7 convenience stores in a row! Now trying to make maps look a bit nicer with mapbox.
[1] https://craftmygame.com/game/convenience-store-marathon
[2] https://craftmygame.com/game/convenience-store-marathon/rout...
Improving the open source automated documentation maintenance workflow I setup https://entropicdrift.com/blog/prodigy-docs-automation/ Any feedback is very welcome.
A Mac-based video manager that automatically transcribes, translates and summarises videos. I process information best through reading, so I built it to manage my growing collection of training course videos, webinars and meeting recordings. Currently working on adding RAG search to make it easier to query content.
Also building a CMS and static site generator that runs entirely client side in the browser. Pick themes, model content an publish to clean HTML. It also makes content available beyond just the browser, eg in a command line TUI.
https://www.sparktype.org
I’m building a typed, array-oriented dataflow compiler that takes small declarative schemas and emits plain Ruby and JavaScript, with a C path. It has a mid-end with inlining, common subexpression elimination, constant folding, dead code elimination, loop fusion, and LICM.
Demo to try it out: https://kumi-play-web.fly.dev/
GitHub: https://github.com/amuta/kumi
On and off working on the Navigation API for Node, Bun, Deno, & as a browser polyfill.
Has 90% test coverage, makes use of web platform tests to verify compatibility, and is in use by some larger companies already with the Navigation API soon to become a baseline in evergreen browsers.
The Navigation API effectively is async state navigations. The likes of React has recently added Navigation API support to make use of the browser reload indicator.
https://github.com/virtualstate/navigation
Along with working on a startup day to day :)
I'm working on Travi, an AI-powered travel companion that helps travelers effortlessly discover the best attractions to check out based on their interests, and experience them through rich, immersive audio narratives.
You can check it out at https://TryTravi.com
I'm working on an unremarkable line-of-business CRUD app, but how are you?
Working on a CLI tool to sync packages between NPM registries. I'll be pushing the code here once its done: https://code.hangdaan.com/foster/kagami
I'm working on a web app that creates easy-to-understand stories and explainers for the sake of language learning. You can listen in your favourite podcast app, or directly on the website with illustrations.
https://infinitepod.app/
Most of the testing so far is English/French/Japanese/Mandarin, but I'm eager to add more languages if anyone is fluent and willing to help me evaluate the text-to-speech.
Still gilding this lily:
https://hackernews.life
I'm working on a poker server to allow holdem bots to play against each other (https://github.com/lox/pokerforbots) and a Pluribus-level bot that plays on it (https://en.wikipedia.org/wiki/Pluribus_(poker_bot)).
Enjoying writing some really fast Zig implementations of hand evaluation and CFR-based solvers.
Building https://ottex.ai - a native MacOS app to solve repetitive micro tasks on a computer.
- Transcribe voice to text (especially useful when you need to explain something to Claude code )
- (soon) select text to instantly Check grammar / Improve writing / change tone of text
- (soon) select text to Translate between languages
I discovered that I have a few 10/20$ subscriptions (grammarly, raycast, wisperflow) that do embarrassingly simple stuff I can one shot with cheap SLM. So I decided to build a one app specialized in small repetitive tasks on computer.
https://www.tirreno.com ~ the open-source security analytics that your application is missing.
Live demo: https://play.tirreno.com/login (admin/tirreno)
Github: https://github.com/tirrenotechnologies/tirreno
I am working on a repository for AI Intent Driven Development at https://github.com/Exadra37/ai-intent-driven-development/.
An Intent is a self-contained document that describes a user request. It is composed of three main sections: WHY (the motivation), WHAT (the requirements, often in Gherkin language), and HOW (a detailed, step-by-step implementation plan defined with tasks). This approach ensures clarity and alignment before any code is written.
Currently working on the first helpcenter that writes itself.
The tool makes it super easy to create help articles in any language, just by clicking through a process. The first results are super promising!
https://happysupport.ai/en
Working on adding Apple Intelligence to my macOS app built to analyze iOS app size metrics. I'm hoping to have a locally running assistant that can act like an iOS build engineer to provide optimization opportunities and more: https://apps.apple.com/us/app/dotipa/id6742254881.
Right now my app allows users to export build metadata as JSON which can be interpreted by LLMs for analysis, but I'd like to have this work on-device.
Data viz experiments with Svelte and Three.js
https://cybernetic.dev
https://www.ottoclip.com - Create product-focused content that stays in sync with your product.
Create a script for a product demo or tutorial for your app using an extension. The script is used to generate your product content in multiple formats (narrated video, interactive demo, looping animation, and in-app guide). Whenever your product changes, just update the script and regenerate everything. No manual re-recording of video, syncing of audio, or any other post-production steps.
New (open source) PostgreSQL index type for analytics workloads, which is a read-only drop-in replacement for B-trees. Smol is multiplicatively faster than B-Trees and radically smaller.
https://github.com/asah/smol
Help, alpha testers, etc all welcome. Sorry RDS/Aurora users: smol is for embedded and self-hosted pg instances only for the foreseeable future.
In the philosophy of selling shovels in a gold rush, I have built a Markdown Viewer for Mac which is optimised for AI coding with the likes of Claude.
It is simple but powerful supporting all formatting but also diagrams so you can get Claude to generate beautiful ER, or state-transition diagrams for your documentation. It also supports math notation, file links and has a cool table of contents feature
It's in the app store: ViewMD
I'm building a scraper in Golang based on Colly to do two things:
* Automatically train the scraper on the structure of the page to acquire the data you want, and
* Clean and structure the data into a format suitable to go into a relational database
I got sick of doing all that manually for some pricing data I wanted to monitor on some suppliers sites, and I've always wanted to contribute more to open source and give back.
I work on Puter (https://github.com/heyPuter/puter/); an open-source, self-hostable internet computer.
People use Puter for an incredibly wide range of things, including cloud storage, web hosting, coding, AI, and gaming. Right now, we're mostly focused on improving performance and making sure that it's as fast as a regular desktop environment!
This looks fun! But the fact that the self hosted version lacks support for some core apps is sad, I'd love to be able to build puter apps! Any plans for an app store-like ecosystem?
Yes! definitely. Please stay tuned :)
I finished my first real attempt at a complete side project as a first year dev.
ytrss.xyz
Convert YouTube videos into rss podcast feeds you can subscribe to anywhere. Also added a YouTube audio conversion api.
https://www.miscbeef.com/octopoda
Multiplayer QWOP-like where you control one leg of an octopus.
I'm further ahead in the development than shown here, hopefully have the finished thing out with support for multiple games within a month or so (would be faster if I didn't have a job lol)
I'm working on a command-line tool for advanced full-text search of written documents. It works in a completely different way than grep, so it can do a lot of operations that grep fundamentally cannot like proximity searching.
I called it Wosp for word-oriented search and print. I released the first functional version a few days ago: https://github.com/atrettel/wosp
AI Copilot for brick-and-mortar retailers (https://fastquery.ai)
Perfect usage metrics. Store staff spends 35hr per week using my software in a ~7 employee per shift setting. No churn.
Bootstrapped past $100k ARR on my own, just onboarded a co-founder.
Congrats! This is amazing!
I'm a writer, non technical, and I'm building Eyeball - https://tryeyeball.com/
Eyeball is a bookmarks app that turns your own saved links into hyper-personalized playlists. It's like having a personal curator in your pocket that sends you a weekly issue of your own personal "magazine" on Sundays.
Working on AI business coach that can help you at different stages of your startup or small business journey. Here is where you can access the business coach: https://ceo.getbeyondx.com
I'm working on a Yelp alternative called Vibehuntr -- just something different to browse venues using Google's API, with a social layer so I can see what my friends like. It's very rough around the edges right now and it might be completely different by next week. It's been a fun experiment in vibe coding on a full stack. https://vibehuntr.io
An LLM-powered 'offline' journaling/mindfulness app that draws on ancient philosophy. Designed it initially to help nudge my own habit along & keep things fresh/interesting every time I sat down for a scribble-sesh.
https://www.gnothi.app
Working on Afterchive, I recently started to my github repo is kinda modest. https://github.com/asemshaath/database-backup-utility
Ultra low-power LoRa stuff with STM32 microcontrollers. Powered by solar.
Oh nice! What protocol are you using?
I’m building Cozy Watch, a macOS app that brings GitHub notifications straight to your desktop.
It tracks pull requests, CI results, and mentions in real time — so you know when you’re needed without checking GitHub or digging through emails.
It has a menu bar for quick access and a clean desktop UI for more detail.
https://www.cozywatch.com
Some years ago, I co-founded a startup that would run workflows when email messages arrived, any email from any source was parsed and it would trigger "actions" that could include notifications - this sounds like a good use case for it! You don't need the service to expose an api to listen in, since most services end up sending email as last fallback.
Thank you for you feedback!
Do you still run that startup?
With Cozy Watch, I use the GitHub API, never thought about using emails as triggers.
I’ve actually got GitHub emails disabled, they can get pretty spammy.
I'm working on an AI-controlled VPS. It makes a lot of sense to let them run with full permissions.
https://www.zo.computer/
There is a lot of repetition when it comes to building AI system. Frameworks don't help. No-code builder are still too rigid.
We are making "batteries included" API to bring agentic AI into any platform.
https://docs.cbk.ai/
https://RadioPuppy.com - Listen to 1000s of online live radio streams.
This is a pet project for myself. I love listening to online radio while at work, helps me focus. But I didn't really click with any of the current selection of web apps out there so decided to build one myself.
It uses the great API available at radio-browser.info for all the radio information.
Been using it as a way to learn how to market a website as well. Learning a lot.
I welcome any constructive feedback.
Very cool, will be using this.
Would suggest that you filter out any radio stations where the URL isn't working if possible.
For example I filtered down to "United Kingdom" and then "bass" - 3 of the 6 worked and would rather see ones that are active.
Also if possible to apply the country filter within the search bar, took me a second to realise I had to open the filter for country, select that, then go back to my search.
When clearing my search of "Bass" in the example above, it reset the search to default (didn't have my country filter) even though the filter was still applied when opening the filter section.
Super easy interface to use though, really well done.
Hey thanks so much for checking it out.
It is annoying when some of the stations don't work. I've done an initial try at filtering them out without much luck but will try again.
Good idea about the country filter.
Thanks for the feedback really appreciate it!
I'm working on https://teeming.ai, trying to solve the information asymmetry problem in the job market.
The project has been a huge learning curve for me - I started out as a skeptic of how generative AI could solve real problems (rather than just create noise) but now think that, like the internet, it can create a new kind of abundance that will be harnessable in all sorts of interesting ways.
A modern QuickBooks, based on beancount, WorkBill (https://workbill.co). You can play with it at https://demo.workbill.co.
Unlike traditional accounting platforms we expose the ledger model directly which enables our customers to model complex transactions even when we do not have direct support for it.
Been working on this for a month, and it uses Elixir, Phoenix and InertiaJS with React.
I'm building a tab manager extension for Chrome, that also kills duplicated tabs.
Why?
The one I used died (Manifest V2 only, and was not updated). And I wanted to test one-shot it.
Incredibly it worked!
I have found that duplicated tabs can be useful e.g. for pages where footnotes are not hyperlinked in the text. When this happens I open a duplicate tab and scroll to the bottom of the page on it.
oh, for sure, that's why the extension shows which tabs are duplicated, and I can kill the duplicates individually, but also has a kill-all-duplicates button
Working on https://www.vantage.sh/
Specifically working on our FinOps agent which can identify and remediate cloud infa cost related issues across AWS, Azure, Datadog, etc. The agent lives in Slack and surfaces cost savings initiatives for teams to inspect and approve for the agent to fix.
I'm building a cursor style ai agent but for planning hikes/trips. It does context management and tool calls into data sources and navigates the world to find interesting places. Should be getting out of private beta this week! https://wanderfugl.com
A FLAC encoder/decoder written in Guile scheme. I struggled to get the decoder working with most test files for a while until recently. It's more or less a fully functional decoder now. It's also 1:1 with the reference meta-flac command currently as well.
https://github.com/steve-ayerhart/guile-flac
A notes app, where you can create flashcards on your notes/pages directly
you can also infinitely nest your notes/flashcard decks, and turn each note into a dedicated page
spaced repetition coming soon
https://studybranches.com
I created this recently but have let it fallow in the last month. Planning to update it over the next few days / weeks. There are a crazy number of directions I could take it.
https://housepriceguess.com/
Would be great to collaborate with others on it. In particular I want to explore building the "alpha arena for AI house price prediction"
https://tailgator.app, playing around with serverless Tailscale nodes
https://coolinary.app, simplifying cooking and recipe ideas
https://capi.tax, preparing capital gains tax reports from foreign brokers for German income tax (still closed)
A filmmaker community for those wanting to showcase their work. Right now everyone's got their own squarespace and the problem is that about 0 filmmakers also want to be web masters.
IMDB? Ads everywhere. Actor's Access? Ancient.
https://cinesignal.com
Human first, AI optional. A great way for actors, writers, and directors to represent themselves.
Feedback welcome! -M@
Working on a location based social app. https://pinggy.com
Ahh, I wanted to build this years ago. Nice one!
A database populated with audio metadata (including a link back to YouTube or Spotify or whatever) that includes vector embeddings for the audio. That way I can grab clips of music I like from YouTube, generate vectors for them, then find similar things in the database.
It's off to a rocky start though, as I've initially populated it with YouTube-8M and AudioSet, neither of which are music-specific. The search results can be... Weird.
Building a set of experiments that explores LLMs visual understanding of your photos to learn about you, especially given the recent learnings from deepseek-OCR. Part of the experiments delve into storing the memories with GraphRAG so they can be effectively retrieved without losing too information.
A mobile game!
https://maspgame.com
Made with Godot and Swift, a casual manic arcade thing where you pop animals in increasingly exotic/banal locations.
https://felko.io/
Hi HN, we’re a Milan-based fintech startup developing FELKO, an AI-powered data platform that helps banks and credit-holders standardize, monitor and act on debt portfolios in partnership with collection agencies!
As a very recent pet project (and pretty much work in progress), Equations Explained Colorfully (KaTeX + Markdown + TypeScript) https://github.com/stared/equations-explained-colorfully
Thinking on good export formats (except of taking screenshots and Pull Requests, obviosuly). LaTeX and Typst? A remark plugin?
I work on Harbor (https://github.com/av/harbor), it is a project to save hours when setting up LLMs locally.
I pre-integrated over 50 different LLM-related projects, added a nice CLI and a Desktop app on top to manage the configs.
I want to make it easier to just quickly enable wake lock on your device in a cross platform, no install, offline capable way. It's a silly little project but I'm super proud of it.
https://wake.lol/
Working on asking whoishiring not to forget the freelancers next month.
https://news.ycombinator.com/submitted?id=whoishiring
Edit: oops nevermind it's a goner.
https://news.ycombinator.com/item?id=45804464
I'm working on integrating a trained nnunet MRI segmentation model (bibsnet) into my cortical surface reconstruction pipeline.
Expanding a speech-to-text (dictation) gnome extension I wrote to work with gnome 49.
https://github.com/kavehtehrani/gnome-speech2text
Working on https://ziva.sh/, an AI agent for game development. It uses MCP to integrate with Godot, a leading open source game engine.
It's coming together really nicely, targeting a beta release later this month. If anyone is interested in game development and wants to be a beta tester, lmk :)
https://michigan-pulse.com
I am building a community driven data aggregation platform for the Michigan tech ecosystem. This is just a promo page.
On launch there will be a company index, curated newsletter, educational resources in michigan like CS programs, and much more!
I signed up. From GR area too. Good luck with this.
Thanks Sam! Means the world. Looking forward to any feedback you have when I launch.
A cpp code generator like esphome, to generate the firmware for midi devices in a simple yaml file, for raspberry Pico.
It would have been so much easy just to program the midi hub I wanted to program but wanted to make it generic.. now I can make the firmware for any configuration in seconds!
https://bgpipe.org/
I'm working on a man in the middle proxy for BGP, which can fix and inspect routing sessions on the fly. Like a firewall for the BGP control plane.
Some time ago I was building a mitm proxy myself, then I found out about: https://www.mitmproxy.org/ Maybe you already had it in your radar
That's for HTTP/S and related, as parent said his is for BGP which is a completely different protocol
Main Project: Radar Analysis Software.
Side Project: Sentiment analysis (±10) on news articles to be used ( along with other indicators ) for stock buy/sell recommendations.
This week I’m publishing my open-source, email based commenting system for websites.
https://r3ply.com
Helping my recent MBA grad sister make a simple python script to fit here resume to a JD using openAI's api. Shes applting to product and marketing roles in AI and this helps her understand the tech (and its limitations) better as well as apply to more jobs easier
Market is brutal though man. She hasnt gotten an offer after so much trying
My latest is Marvelogs (https://www.marvelogs.com) - always wanted to build a price tracker (or tracking any values on a regular basis) - it's nearly there.
How does this work? Good old scraping/parsing html or you feed the html to a LLM and let it parse?
Nice idea btw!
system to test and calibrate an analog traction control system. the system uses a frequency to voltage converter and a bunch of opamps to compare wheel speeds then determines wheel slip or slide and either reduces engine power or braking.
Test system uses ADCs, DACs and a DDS to produce a sine wave that simulates wheel speed.
I would rather be fishing.
Building a tourist app for my local city here upstate New York, since nobody in the local government office can be bothered.
Good luck! It can positively impact the whole city, by a lot.
A DeepWiki to mdBook converter to automatically document some other projects I'm working on: https://docs.deepwiki-to-mdbook.zenosmosis.com/
I wrote a pretty complicated set of GNU Makefiles for a simulation library at work, but was annoyed I had to work so hard to avoid collisions, so I'm working on a "more sanitary" build-your-own-build-system/build-system-kernel type deal.
Working on a Document Manager in the Apple Ecosystem.
https://apps.apple.com/us/app/document-manager-fyle/id674003...
Evals for programming languages with formal verification. It's not clear how far we are from good coding performance in less popular languages in general, and formal verification has some quirks on top also.
I'm building a Firefox, Chrome, VSCode and OpenVSX security scanner and profiler, and working on building a private web store for Enterprises to switch to rather than using the default stores given all the ransomware and malware activity in that space. Will show HN very soon!
Recently working on https://rainsounds.xyz
Here you may mix rain, thunder and more sounds.
An AI development workflow platform with GitHub integration. Built in Elixir / Phoenix. Early stages but it's a fun project.
https://github.com/kenforthewin/matic
I am working on a web app that doctors/NPs/PAs can use to automatically rewrite complicated and verbose medical progress notes. The amount of time medical providers spend on documentation is ballooning, and only a small portion that time is actually spent doing medical decision making. The rest of the time is spent incorporating (ie copy editing) data as it comes in from imaging/bloodwork/consultant advice. The goal is for the app to be:
entirely self service, without needing EHR integrations
able to persist and reuse the user's writing style, without actually saving any of the notes
HIPAA compliant (obviously)
No projects at the moment. Just working on myself and improving some things in my life, job, cheaper place to rent, lose weight etc. Dreaming of starting a business, I just want to add a cool service to my local city but the economics is hard
Just got a 3d printer (Bambu a1 mini) and my girlfriend brought home a whole bag of plant cuttings. Thought I would give a modular plant pot (i.e with elements that allow for expanding the pot) in fusion 360 a shot.
I'm working on MedAngle, the world's first agentic AI Super App for current and future doctors. Invite only, 100k+ users, 150m+ questions solved, tens of billions of seconds spent studying smarter.
I'm reworking my dungeon generator on Iron Arachne as a result of what I'm studying in Applied Algorithms for my master's degree.
More deterministic, much improved time complexity, and hopefully, more interesting results.
I’m working in a universal UI for agentic coding in the terminal.
https://willmcgugan.github.io/toad-report-2/
Basically small data tools for myself.
A non-bloated HTML, CSS and pure Vanilla JS framework to create dashboards.
A cross-platform JSONL viewer where I am learning ImGUI. Haven’t found any other open source GUI framework that‘s small, provides out of the box components for tables, sorting
https://kintoun.ai
Document translator that keeps layout intact.
I'm working on a simple open source iOS client for hacker news that makes use of the latest iOS design language and features.
No Ads, no paywall, just focused on a good reading experience with some extra niceties like widgets on the home screen.
Website: https://www.hackerreader.app/
App Store: https://apps.apple.com/us/app/hacker-reader/id6754137305
Repo: https://github.com/danielcspaiva/hacker-reader
https://getfast.ai/strava/login Using transformers to understand fitness data
Onchain consumer credit (x402 “credit card”)
https://x.com/philip0x/status/1982219251479097601?s=46
Building an ai powered threats deception proxy.
It's a honeypot system that uses AI to mess with attackers. When someone tries to hack your app, it detects them and serves up fake responses based on attack type.
The system learns from attackers behavior and creates convincing decoys to waste their time and frustrate their efforts. It's basically a trap that gets smarter the more attackers poke at it.
https://github.com/0tSystemsPublicRepos/IfritProxy
Selfhosting via yunohost, especially loving Immich and Actual Budget. I'm not in IT but I use Linux and it has been easy enough for me to set up.
Also making personalised Christmas t-shirts in Inkscape. I love what you can do with open source tools!
Trying to streamline pet travel compliance and making sure nobody gets stuck at the border with the help of AI and some veterinary partnerships.
https://superpets.app
I am building a mobile podcasting app that uses ML to auto-skip ads
Been nerd sniped recently so am working on a Rust version of markdownlint-cli2. I'm tired of having a node dependency in my projects and this seems like a constrained enough problem space that I'll actually get around to doing it.
We're working on https://www.octozoo.com - helps development teams improve their code quality by tracking code coverage.
Also planning on adding more tools to help development teams.
A template-based automation tool for small private equity firms. It has some AI functionality as well to easily parse documents and information from transcripts. Basically I want to free up investment teams from admin tasks so that they can spend more time on evaluating deals and building relationships.
A lot of the AI-powered applications for private equity firms are focusing on the multi-billion dollar firms.
https://www.unsilodata.com
8086 assembler in awk for retrocomputing purposes. Proof of concept worked well enough and now i'm doing an second version with more robust logic.
Trying to teach LLMs to speak proper https://en.wikipedia.org/wiki/Lojban.
systemg - "Systemd, for busy people".
https://sysg.dev
https://github.com/ra0x3/systemg
I'm personally tired of getting stuck in config/deployment hell every time I want to deploy a long-lived web service. Sure I eventually learned how to use systemd, but systemd has SO many things baked into that I simply don't need. systemg is a lightweight process supervisor that features everything you'd typically want when running/managing production web services in the wild.
Would love feedback.
I love that it's Rust based, but being busy is exactly why I like Systemd, it Just Works, as long as you don't need to customize the OS at all.
Working on a mobile astronomy app that uses AR to determine your unobstructed view and predicts what you can see and when
Anti traffic tickets system in brazil, booming market, reply if you're interested, looking for experienced partner or angel.
Moved from nodered only to a hybrid of nodered and home assistant. Added some new sensors, nfc tags, modes and automations for multiple tenants / cost savings. Its been fun to automate some boring tasks.
Using GenAI to build small highly useful tools. Poker tracker, calorie tracker, budget tracker.
Is anyone working on or knows a library for evaluating LLMs for application features and/or application features that use LLMs? I am wondering what people use or if anyone has their own solution.
There would be so much subjectivity to this. I like the idea but executing in a reliable, repeatable way would be very challenging imo.
https://prepbook.app - minimal recipe manager
As simple to use as a notes app, with clever culinary capabilities :)
Would be great to be able to choose options for metric/imperial, fahrenheit/celcius and spoons&cups/ml&g.
I’m building Streamlit for Java. https://javelit.io https://github.com/javelit/javelit
I maintain a dev log: https://world.hey.com/cdecatheu/javelit-diary-00-building-a-...
And here’s an article about the project by a Google Cloud devrel :)
https://glaforge.dev/posts/2025/10/24/javelit-to-create-quic...
Making my first ATE (automatic test equipment) and considering whether I should use diy linear power supply or buy dc-dc switching module.
decentralized bluetooth gossip network ios and android app auraphone https://www.youtube.com/shorts/J-zxK7Xl6z0
I'm working on an invoice parser for small businesses to use. It's a fun project but the integration is going to be a bitch.
Carimbo, my 2D game engine
https://github.com/willtobyte/carimbo
An overengineered plant watering machine using an old RPi to get better at basic electronics.
I'm adding a Shortcuts‑like UI with Hyperscript syntax for defining logic to the app builder I'm building.
Grugnotes.com still my personal project after a few years. And lately how to make agents useful for notes.
Just learning. Interested in prolog.
I'm working on some writing on my blog [1], trying to improve my writing and explore style.
My current series of post follows the surge of interest in UUIDs with the uptake of UUIDv7. I've seen some subtle misunderstandings spreading, so I dive into nuance. This has spun off some mini projects, like an RFC compliant UUIDv8 implementation based on XKCD 221 [2] (humor intended). I think I have two more in the blog series.
[1] https://alexsci.com/blog/uuid-oops/
[2] https://github.com/robalexdev/uuidv8-xkcd-221
I am building a coding agent for small businesses. The agent runs on Linux box on own cloud. Desktop and mobile apps to chat with AI models and generate software as needed.
SSH based access with HTTP port forward. Team collaboration, multiple models, git based workflow, test deployment automation, etc.
Very early stage but it now work on its own source code (Bash tool is missing): https://github.com/brainless/nocodo
Not an original idea, but I’m working on writing a minimal OS to run Doom.
Building an ai powered threats deception proxy. A dynamic honeypot system that uses AI to mess with attackers.
When someone tries to hack your app, it detects them and serves up fake responses designed to make them think they're getting somewhere.
The system learns from attackers behavior and serves ai generated decoys to waste their time and frustrate their efforts.
It's basically a trap that gets smarter the more attackers poke at it.
MVP version at https://github.com/0tSystemsPublicRepos/IfritProxy
Brilliant.
alapaca papaer trading automated on a raspberry pi - algo trading is there for a while and this would help me test my strategies
ArtCraft [1], a source available Adobe + AI.
Here's a demo / trailer that shows it off:
https://www.youtube.com/watch?v=NNHSTfWbkaA
If you're into movies or filmmaking, it's a fantastic AI tool for consistent, fully-intentional scenes with deliberate set and actor blocking.
It's also the cheapest model aggregator service out there. You can log into every AI image and video provider directly and don't have to pay me anything to use the tool. You can use your Sora account, Midjourney account, Grok account, etc. It'll soon let you log into other aggregators like OpenArt, plug in your FAL API key, etc. so you can use your credits/funds wherever they happen to live.
Unlike the other "model aggregator" websites like Higgsfield, this is a desktop app written in Rust that you can keep. It also has highly intentional 2D and 3D design surfaces especially built for design.
Text prompting sucks for artists and designers, so I'm trying to put image and video design onto canvases that you can intuitively mold like clay.
Here are some short films made with ArtCraft:
https://www.youtube.com/watch?v=H4NFXGMuwpY
https://www.youtube.com/watch?v=tAAiiKteM-U
https://www.youtube.com/watch?v=SuVW8l-_O3I
Would love to hear feedback if anyone tries it out.
[1] https://getartcraft.com
Real time guitar effects simulator
I am working on my website https://moverstoo.com/
I’ve recently written ImapGoose, a daemon which keeps a remote IMAP mailbox in sync with a local tree of Maildir: https://whynothugo.nl/tags/imapgoose/
It relies on “modern” (2009) extensions to minimise traffic and avoids polling entirely (relying on the server to notify of new messages or changes as they happen).
It’s currently quite stable. The only known issue is that it can take a while to detect a timeout when the system is suspended and woken up again (there’s no portable API to detect suspend/resume).
Since then, I’ve been working on a simple TUI email client based on notmuch and maildir. So far it works really well for processing email, but lacks any capabilities for handling attachments, composing, sending (these are obviously on the roadmap).
Oh I'm super curious about your notmuch project. I've been wanting an email client that's just a tui on top of notmuch. Wonder if we can pair up? Ill ping you over email
I just finished a little webtoy. It's like a comic strip time machine - you can see a virtual newspaper comics page for any date in the last seventy years.
https://lmao.center/funnies
V happy with how the CSS came out, except I spent a lot of time on an "ink bleed" newsprint effect that (oops) only looks good on HiDPI monitors... lessons learned I suppose
A JVM for Java 21 written in go[0]
[0] jacobin.org
https://xelly.games/
Users post small games to social feeds.
Scroll like a social network, jump into and play any game by tapping on it.
Games are served into fully locked-down, sandboxed iframes for security.
Docs for game developers: https://xelly-games.github.io/docs/intro
I buy and operate e-commerce brands that sell on Amazon, and I'm working on handing as much of the operation of the business off to AI as possible. Doing this both for actual time savings for myself and also as my big-picture eval of new AI models + products as they come out.
I also started a Substack to document it - here's a recent post on using Gemini to screen inbound emails with prospective acquisition targets via a Google Apps Script that evaluates the listings in those emails daily: https://theautomatedoperator.substack.com/p/screening-inboun....
https://gametorch.app/sprite-animator
Create video game sprites and animations via prompts.
Pretty excited because I've started to get high volume, repeat customers.
Ahhh this is exactly what I'm looking for! I don't see any pricing on any pages. Would love to know how much this costs (I don't know what 455 diamonds is worth) as there's a few sprites that I'd love to animate and use in my app.
Not a fan of signing up before seeing how much I'd have to pay. The examples look great though.
1 diamond = 1 cent
Prices are about to drop dramatically. Many of the models dropped >80% in price since initial launch. Any time I have a reduction in cost, I pass the savings directly on to users.
NativePHP
Hypertufa plant enclosures
Added compression to my metal 3D printer slicer exported CAM files, and refactored the code to better support larger volumes.
Tweaking the piezoelectric driver PCB design for the micro-positing microscopy stage project. The Nanomotion piezoelectric motors were not meant to be used in the manner I chose, but it is fun to push the limits of technology.
Finishing up some custom 1U mounted hardware, and getting a batch of test PCB soon. Bend radius came back 1mm oversize, but this was acceptable for a single run item.
Also involved in several other projects maybe 3 people would care about. Doing a custom FPGA PCB is not very fun unless encountering that rare class of problem CPU/MCU simply can't handle cleanly. =3
Sewing my own clothing.
But of course the programmer in me needs to make my own software to design patterns with code. Enjoying using paper.js to do all the complicated math to calculate lengths and angles.
https://homefree.host
All-in-one router/nas/firewall/adblock/app server (each piece optional)
Declarative and reproduceable as it is built off of NixOS, but administered through a UI, so the user doesn't have to know this.
All state managed in a backup bundle, so it can be hosted at home or in the cloud.
Goal is to have a box you plug just like a wifi access point into your modem, follow a simple web-based installation flow, then you are running a personal cloud.
Website is self-hosted by HomeFree, but installation instructions are very out of date, which I'm working on right now. There are now installation ISOs that I will soon add a link to.
I've been building SageNet, a voice-first AI coach that turns your goals into structured, adaptive learning plans.
After a 2-minute voice conversation, Sage generates a personalized 6-module roadmap with build-first projects. It checks in by voice, analyzes your reflections, and regenerates your plan if needed. You can invite friends to your Support Squad for accountability.
The biggest insight so far is people don’t want “infinite content.” They want structure and someone who remembers them.
would love feedback!
http://sagenet.club
nunya
In network code: most people just let the OS choose a default adapter. It works fine, but it makes it hard to write software that works across machines with either (1) multiple NICs (and/or networks they point to.) or (2) multiple external Internet IPs. Look at STUN, for example.
A STUN server that lets people test what type of NAT they have uses two IPs. For such a server you have to manually specify the addresses to bind on to make for sure its setup right. As it goes, writing network software to do simple things like "bind on all local addresses", "bind publicly", "bind on all", is harder than it sounds. There are edge cases on different OSes and address families, so manually managing IPs is hard to do.
My network software lets devs easily manage NICs and routes they support without guessing about addressing. Additionally, I've written a bunch of software with the library already to do things like NAT traversal. So its really my own redesign of how to do networking on the Internet. Designed to hide a lot of the messiness. I'm still improving code quality so it's not ready yet. But I've been dog fooding with a lot of software written in it and smashing bugs every day.
Project page: https://github.com/robertsdotpm/p2pd Built this recently with it: http://ovh1.p2pd.net:8000/servers (server monitor for public STUN, TURN, MQTT, and NTP servers. Only checks every 4 hours to avoid spamming them though.)
For fun have been creating a mashup of old school DnD map generation using Commodore "10 Print Chr$(205.5+Rnd(1)); : Goto 10" style logic (in TS/Svelte/SVG):
https://imgur.com/a/qMeEoPK
Have been down a rabbit hole ensuring the stairs are realistic and that grid connects properly. Lots of fun and frustration with AI coding tools trying to solve that (they mostly don't/can't). Some fun detours learning a little Prolog to help out as well.
Manabi Reader: Japanese learning through reading
https://reader.manabi.io
I recently added FSRS (besides also having Anki integration). Now I'm working on replacing the need for reviewing flashcards by having reading activity automatically mark flashcards (current and future) as reviewed, so that you can get many of your reviews in just by reading native materials that interest you instead of sacrificing most of your study time to contextless flashcard grind.
I'm also working on a manga mode using a new manga OCR tech I have licensed out of academia that is ahead of state of the art alternatives.
This project now sustains my full-time focus.
This looks super cool. I love to see people working on language learning tech. I'm working on a language learning app too but it only really works well for indo-european languages. That said I would still love to collab or talk shop, my contact info is in my bio
Here’s mine:
I’m building a small live NFL game-prediction tracker and writing up what I learn as I go:
https://michellepellon.com/portfolio/nfl-game-predictions
# What’s under the hood today
ELO translated to the NFL with margin-of-victory adjustments, a modest home-field term, and week-to-week recency weighting.
Post-hoc calibration with isotonic regression so 70% predictions land near 0.70 empirically.
Monte Carlo to roll games forward for distributions on weekly win odds and season outcomes, plus basic reliability/Brier/log-loss tracking.
# Where I’m taking it (ensemble ideas)
Blend a few complementary signals: (1) pure ELO strength; (2) schedule-adjusted EPA/Success Rate features; (3) injury/QB continuity and rest/travel effects; (4) a small “market prior” from closing lines; (5) weather/play style pace features.
Combine via a simple stacked model (regularized logistic, isotonic on top), or a Bayesian hierarchical model that lets team effects evolve with partial pooling.
Separate models for win prob vs. expected margin, then reconcile with a consistent link so the two don’t disagree.
Emphasis on calibration over leaderboard-chasing: reliability diagrams, ECE, PIT histograms, and backtests that penalize regime drift.
# Why I’m doing it
It’s a sandbox to teach myself Monte Carlo and ELO end-to-end—data ingest → feature plumbing → simulation → calibration → eval—on a domain with immediate feedback every week.
# How this connects to my day job (healthcare ops)
I work at BlueSprig, running ~150 ABA therapy clinics. I’m exploring whether ELO-like ideas can augment ops decisions:
“Strength” ratings for clinics, care teams, or scheduling templates based on outcome deltas and throughput (margin-of-victory ≈ effect size/efficiency).
Opponent/schedule ≈ case-mix, payer mix, staffing constraints, geography.
Monte Carlo for expansion planning (new-site ramp curves), capacity/OT forecasting, and risk-adjusted outcome monitoring with calibration so probabilities mean something.
Guardrails for fairness and interpretability so ratings don’t become blunt scorecards.
# Help
If you’ve shipped calibrated ensembles in sports or have pointers on applying rating systems to multi-site healthcare operations, I’d love to trade notes or if you need someone to this and other kind of work for their dayjob email me at mgracepellon@gmail.com -- I would love to do this fulltime.
LLM Agents.
Training or integration?
More of integration
https://limereader.com/
A time-sorted list of top posts from Hacker News, Tildes, Lobsters, Slashdot, Bear, and some science, tech & programming related subreddits.
Posts on STEAMD topics (Science, Technology, Engineering, Arts, Math, and Design).
My site went live 2 days ago. I shared more details on below post but for some reason, my post was shadow banned and didn't show up on Show HN.
https://news.ycombinator.com/item?id=45849924
Any constructive feedback is welcome!
Note that I am trying to narrow down a bug in my backend which sometimes causes it to crash. Since backend is built in Swift using SQLite as database, it's a bit hard to nail down the issue.
A library of components for Godot that could be used in different kinds of 2D games: https://github.com/InvadingOctopus/comedot
Dreaming about a new programming language made for coding gameplay logic: https://news.ycombinator.com/item?id=45865379
And an iOS expense tracker focused for frequent travelers, and macOS photos viewer based on the filesystem instead of a monolithic opaque "library", 2 needs that I had since forever but could never get through Apple's atrocious developer documentation far enough to finish making them :')
A lot of things!
A law professionals helper - aggregates judicial case info into a single place, gives visibility and notifications - asistentul.ro
A scheduling platform for self-employed professionals that offer services (think hair-cutting, nails, psychlogists). (Not yet live)
Aaand something in compliance that I want to keep a bit stealthy right now.