I wanted to make JSON/YAML configuration language for my projects. And i wanted a strict specification. This is want i created, now with specification and 100% coverage, reference implementation it’s just one prompt to reimplement parser in another language.
I wanted to make it easier to quickly see/study trending articles on Wikipedia because they tend to make good topics to know before going to trivia night.
I've had the domain for awhile, but just made the app today on a whim.
I use Wikimedia's api to get the trending articles, curate them a bit, add some annotations to provide some context, then push to deploy the static site.
After winning the Playlin Player's Choice award I've noticed an uptick in players as well as some people sharing videos on YouTube which has been fun. I've got a few thousand people playing every day.
I just launched user accounts today so user's can now track their progress across devices and share their stats with each other. This ended up being a bigger chunk of work than I expected but I'm really pleased with how it turned out. (Though I launched it 15 minutes ago so I'm holding my breath for bug reports)
I'm fine-tuning my internal puzzle-building now with the goal of letting people use them to make and share their own puzzles soon!
It's an iOS app that applies various generative art effects to your photos, letting you turn your photos into creative animated works of art. It's fully offline, no AI, no subscriptions, no ads, etc.
I'm really proud of it and if you've been in the generative art space for a while you'll instantly recognise many of the techniques I use (circle packing, line walkers, mosaic grid patterns, marching squares, voronoi tessellation, glitch art, string art, perlin flow fields, etc.) pretty much directly inspired by various Coding Train videos.
Hey, I've been getting into visual processing lately and we just started working on an offline wrapper for Apple's vision/other ML libraries via CLI: https://github.com/accretional/macos-vision. You can see some SVG art I created in a screenshot I just posted for a different comment https://i.imgur.com/OEMPJA8.png (on the right is a cubist plato svg lol)
Since your app is fully offline I'd love to chat about photogenesis/your general work in this area since there may be a good opportunity for collaboration. I've been working on some image stuff and want to build a local desktop/web application, here are some UI mockups of that I've been playing with (many AI generated though some of the features are functional, I realized that with CSS/SVG masks you can do a ton more than you'd expect): https://i.imgur.com/SFOX4wB.pnghttps://i.imgur.com/sPKRRTx.png but we don't have all the ui/vision expertise we'd need to take them to completion most likely.
I'm working on a fully offline, client-side train journey planner for UK rail - https://railraptor.com
When booking flights, I use sites like Kiwi and Skyscanner that let you do flexible searches - multiple destinations, custom connections, creative routes, etc. But rail search feels oddly constrained. All the UK train operators offer basically the same experience, and surface the exact same routes. I always suspected there were better or just different options that weren’t being shown. Where is the "Skyscanner for trains"?
After digging through the national rail data feeds, I decided to have a go at building my own route planner that runs completely offline in the browser. This gave me the freedom to implement more complex filters, search to/from multiple stations, and do it without a persistent network connection.
Now I'm finding routes that aren't offered by the standard train operators, connecting at different stations, and finding it's often easier to travel to different stations (some I'd never heard of) that get me closer and faster to where I actually want to go!
It's still a little rough and I'd like to add more features such as fares, VSTP data, and direct-links to book tickets, but wanted to share early and get some initial feedback before investing more time into it. So, thanks in advance - let me know what you think.
This sounds very nice! A slightly adjacent question: have you discovered any providers that can recommend train journeys based on price? Sort of like the explore feature you find on sites like Google Flights, Ryanair and Flixbus. Sometimes when the wanderlust hits I've tried searching around for cheap train tickets, but it isn't simple using sites likes DB/OEBB/SBB/SNCF/etc
My goal is to make a simple yet interesting procedural and replayable puzzle. It has a couple of weekly variations: on Saturdays you need to break a rule to score max points, and on Mondays there's an added memory aspect which brings variety to the game.
It's mostly vibe-coded which lets me focus on game design and testing. The next step is better onboarding/tutorial and more intuitive UI.
Thanks for trying it out. I'm experimenting with some more variants, for example having more 'rules' than cells, so that you have to choose which ones you'll use.
I'm working with Claude Code to create complete programming systems in languages other than English. Not just wrappers around an English syntax;these are based on an English original but are complete scripting languages in their own right, with documentation, tutorial and programmer's playground. Each variant has its own language pack and they share a common compiler and runtime. The best of all is they are extremely AI-friendly. I've started with Italian and I'm looking for collaborators to work on others. I'd like to do Polish and Bulgarian but any are possible. See https://allspeak.ai.
It's a native Clojure dialect which is also a C++ dialect, including a JIT compiler and nREPL server. I'm currently building out a custom IR so I can do optimization passes at the level of Clojure semantics, since LLVM will not be able to do them at the LLVM IR level.
I'm working on https://concludia.org, a site that helps groups of people collaborate on arguments and conclusions. I don't really have any revenue plans for it currently as I suspect it will be rather niche -- I certainly wouldn't mind if it tops out as a small community of users -- but I've found it super useful in various contexts at work and at home.
You can read more about it over at the site, but it allows you to construct and validate arguments in a graphical form, and it has truth/proof propagation so you can see whether a conclusion is currently considered valid or contested. Some upcoming plans are to allow users to validate arguments for themselves, like mark which parts they understand and agree with so they can collapse that part of the graph, and to add more mcp capability so that LLM can help you construct and validate new arguments.
Inspired by Ralph loop and bash scripts, I created my own version of it where I focus on finding code issues and auditing my codebase. It runs N iterations after mapping the whole code.
I'm working on https://suggestionboard.io, a live polling/feedback/Q&A webapp that doesn't require an account. Just launched the first version, now looking at how people use it and other similar apps and making small improvements.
I'm back to searching for numbers that are palindromes both in decimal and in binary. [0]
I had an insight the other day, that as I fix the n least (and most, it's a palindrome!) significant decimal digits, I also fix the remainder from division in 5^n. Let's call it R. Since I also fix by that point a bunch of least (and most) significant bits, I can subtract how much they contribute mod 5^n from R, to get the remainder from division in 5^n of the still unknown bit. The thing is, maybe it's not possible to get this specific remainder with the unknown bits, because they're too few.
So, I can prepare in advance a table of size 5^n (for one or more ns) which tells me how many bits from the middle of the palindrome I need, to get a remainder of <index mod 5^n>.
Then when I get to the aforementioned situation, all I need to do is to compare the number in the table to number of unknown bits. If the number in the table is bigger, I can prune the entire subtree.
From a little bit of testing, this seems to work, and it seems to complement my current lookup tables and not prune the same branches. It won't make a huge difference, but every little bit helps.
The important thing, though, is that I'm just happy there are still algorithmic improvements! For a long while I've been only doing engineering improvements such as more efficient tables and porting to CUDA, but since the problem is exponential, real breakthroughs have to come from a better algorithm, and I almost gave up on finding one.
Published 3 articles so far, but working on AI architecture and management. While most people are focused on prompt engineering and making stuff with AI; I'm more interested in how it actually works, how to size workloads, how to maximize performance, the security and safety aspects. Here is my most recent article where I played with benchmarking tools to get a baseline and understand how configurations impact token generation
I believe that AI-powered software development means we need to fundamentally rethink how we preserve code quality.
Model output volumes mean that code review only as a final check before merge is way too late, and far too burdensome. Using AI to review AI-generated code is a band-aid, but not a cure.
That's why I built Caliper (http://getcaliper.dev). It's a system that institutes multiple layers of code quality checks throughout the dev cycle. The lightest-weight checks get executed after every agent turn, and then increasingly more complex checks get run pre-commit and pre-merge.
Early users love it, and the data demonstrates the need - 40% of agent turns produce code that violates a project's own conventions (as defined in CLAUDE.md). Caliper catches those violations immediately and gets the model to make corrections before small issues become costly to unwind.
It has some interesting applications for building high performance clients for mssql with tds protocol implementation.
The APIs allow almost direct data serialization to wire instead of datatype materialization in rust.
Makes for a suitable contender for high performance language interop.
I'm writing an essay about how I use an ancient text editor, GNU Emacs, along with gptel, Gemini, some local models, yt-dlp, and patreon-dl to help me me study an ancient language, Latin.
I want to show how I liberate poorly aligned, pixelated PDF image scans of century-old Latin textbooks from the Internet Archive and transform them into glorious Org mode documents while preserving important typographic details, nicely formatted tables, and some semantic document metadata. I also want to demonstrate how I use a high-performance XML database engine to quickly perform Latin-to-English lookups against an XML-TEI formatted edition of the 19th century Lewis & Short dictionary, and using a RESTXQ endpoint and some XQuery code to dynamically reformat the entries into Org-mode for display in a pop-up buffer.
I intend demonstrate how I built a transcription pipeline in Emacs Lisp using tools such as yt-dlp and patreon-dl to grab Latin-language audio content from the Internet, transcode the audio with ffmpeg, do Voice Activity Detection and chunking in Python with Silero, load the chunks into Gemini's context window, and send it off for transcription and macronization, gather forced-alignment data using local a local wav2vec2-latin model, and finally add word-level linguistic analysis (POS, morphology, lemmas) using a local Stanza model trained on the Classical corpus.
This all gets saved to an an XML file which is loaded into BaseX along with some metadata. I'll then demonstrate some Emacs Lisp code which pulls it into an Org-mode based transcription buffer and minor-mode for reading and study, where I can play audio of any given Latin word, sentence, or paragraph, thanks to the forced-alignment and linguistic analysis data being stored in hidden text properties when the data was fetched from the database.
Lastly, I'd like to explore how to leverage these tools to automatically create flash cards with audio cues in Org mode using the anki-editor Emacs minor mode for sentence mining.
I've been writing about interesting books and papers I read for a few years now. I wanted a nice, simple interface to point people to as a "hub" for recommendations that's compatible with a static site.
https://github.com/Realman78/Kiyeovo - I'm currently working towards the full release of my P2P dual-network mode messenger which is currently in beta. The reviews were overwhelmingly positive when I released the beta a week ago so that motivated me to try extra hard to make it pseudo-perfect upon full release
I'm super proud, because it came to my knowledge that someone at Codex used my tool to debug codex+zellij issues, by running zellij within `tu`, and then codex inside zellij
I’ve been building quantum photonics experiments. Repeating the Bell inequality tests that won the 2022 Nobel, quantum erasers, etc.
I just published a fun interactive 3D demo of SPDC, one of the most common and accessible ways to create entangled pairs of photons. I'm hoping to publish a series of articles on other cool learnings about doing quantum photonics in the lab.
I have been making GTK applications so that people can manage MergerFS and LUKS encrypted hard drives without knowing how to be a sysadmin.
The use case is kind of neat. RAID is great and can mostly solve these problems, but people don't have SATA hardware that can handle the workload well, plus they aren't ready to manage an array like that, and they don't like having to use specific sized drives, etc. Another major issue with those setups is you need to be careful because an IO error that you don't recover from will be very difficult or impossible to recover from because of the layers of LUKS combined with LVM.
With MergerFS you just use regular file systems that are separate, but they get combined into a single mount point. That means each disk can just be a different LUKS encrypted drive and if you need to recover data it's isolated to that one disk and much more manageable. You can also take any disk and plug it into another machine as needed and grow or shrink the storage pool as needed.
MergerFS has options and settings to help you determine how files are spread across the drives, such as least space used or which disk has the most of that directory path already.
My app (Chimera) automates the unlocking of the disks, mounting and some data migration if you want to remove a disk from the pool. I plan to add some rclone features to help provide easier backup options to places like Backblaze, AWS, or a remote server in general.
So far so good and I was surprised at how well Opus had been handling Gtk and pkexec.
Let me know if you guys are interested I am close to pushing some RPMs and DEBs, in addition to the standard Python stuff.
https://motionparty.net/ - A collection of games you control by waving in front of your camera, similar to playstation eyetoy back in the day. It supports 1-4 players.
I think it works quite well so far, but need to tweak the camera algorithm a bit to make the buttons work better. Thinking about more games to add as well.
I built a cooking timer that solves the mental arithmetic of roasting multiple things at once. Pick chicken legs, sweet potato, green beans, etc and it'll give you a simple plan telling you when to put things in, flip them and take them out. Trying to eat more veggies and home cooking so this has helped me a lot!
I'm building free immigration software for DIY applicants [1]
It's a free USCIS form-filling web-app(no Adobe required). USCIS forms still use XFA PDFs, which don’t let you edit in most browsers. Even with Adobe, fields break, and getting the signature is hard.
So I converted the PDF form into modern, browser-friendly web forms - and kept every field 1:1 with the original. You fill the form, submit it, and get the official USCIS PDF filled.
I found out SimpleCitizen(YC S16) offers a DIY plan for $529 [2]
So, a free (and local-only) version might be a good alternative
I’m making Bezier, a mac-native vector design app as an alternative to Figma and Sketch.
Unlike those apps it has full support for design tokens and (so far) flexbox layouts. It can also export directly to HTML, rather than a fake preview mode. I’m also working on full code-backed components, so you can go between code and design very easily.
As a designer, I’ve been frustrated for years by the gap between design and code, and despite all the new AI features, Figma still hasn’t got any further in years - design tokens need a 3rd party plugin and responsive designs are a pain in the bum. So I decided to build something that has the ease of Figma while being much closer to live code.
I’ve got to the point where I’m designing the app in itself, tokens are working, html export is working and nearly ready for first betas.
PSA: This is the best place to collect upvotes for your vibe coded ideas/projects that you think might not be up to "Show HN" quality yet, whether reproducible at the source code or prompting level(s) of software development or not... the bar is understood to be much lower here.
I've got demos up and running (mirroring/extending PDB and GEO). Next I'm working on APIs with good AX, ML-friendly export, and an unified AI-driven UI that works for all scientific data types.
I'm working on LookAway, a Mac app that reminds you to take breaks from the screen at the right moment instead of interrupting you at random. https://lookaway.com
Right now I'm focused on the stats side. It already shows how much time you spend in each app, and I'm adding website tracking too, which should make the picture much more useful.
I'm also working on better break timing for dictation. LookAway already delays a due break if you're in the middle of typing, so it does not interrupt at a bad time. Now I'm trying to extend that same behavior to dictation as well, which turns out to be a pretty interesting detection problem because it overlaps with some of the other context signals I already use.
Most of the challenge is making it smarter without making it feel more intrusive.
I'm beginning to homeschool my kids in computing, and we are pairing up chapters of The Elements of Computing System (the Nand2Tetris book) with games that teach similar skills/kinds of thinking (Human Resource Machine, Comet 64, etc...), but we didn't find anything to supplement the first two chapters (where you build basic chips up to an ALU in HDL). I ended up starting creating a kind of browser based kata for those chapters here:
Writing. I publish one long form essay a month with two published thus far. The third one is in editing stages. An enjoyable experience moving from internal notes to outward expression.
I'm working on a local desktop app for inventory and production management: https://kitted.site
It includes bill of materials, purchase/production orders, "can I make n?", stock takes, multiple stock locations, and barcode scanning. It's aimed mainly at small business and makers for the time-being, but still allows multiple users to connect over the the local network.
I coded a visual novel/adventure game framework in pygame. Pretty much just to see if ai could handle a full project.
Eventually I got scope creeped into a full game with branching stories, item crafting, and a custom cutscene engine...even Trained a model for a few specific art assets.
I'm working on a digital waveform viewer for VScode. I started it back when I used to work for an FPGA company, and needed to debug soft CPUs. Now it's starting to rival the proprietary software. I should probably do a show HN at some point...
Makes reading/searching the Postgres mailing lists easier.
I’m polling a Fastmail inbox to nearly instantly receive and ingest messages. Anyone can browse without an account, but registered users can follow threads to be notified of new messages, threads in which your registered email is found are auto-followed, and there are some QOL settings.
Search is pretty naive right now (keyword on subjects) but improved search is the next big thing on my list.
Music player, organizer, discovery tool that will load history and subscriptions from streaming services and discogs, last.fm etc and allow you to query it with AI.
A service summarising and simplifying EU laws, resolutiins, decisions and so on: https://euforya.eu/
One thing I find especially intriguing is how LLMs can help deal with desinformation:
- I experiment with deterministic settings of local LLMs for the document summary so that sharing a prompt would prove that the output was not tempered with (no desinformation on the service side)
- I add outputs of several LLMs (from the US, the EU and from China) for the "broader context" section so users could compare the output (no desiformation on the provider and model side)
I'm tired of all the recents npm packages supply chain compromises, so I've written a collection of `sandbox-exec` rules to wrap all the `npm install` and `npm run <script>` of my projects on my machine. It works but it's messy, so now I'm working on a small rust tool that acts as a wrapper and generator for that so that it's nicer to use and can be shared to other people.
I evolved an rsync based backup script I've been using for almost a decade into https://github.com/nickjj/bmsu. I use this for backing up my life's work to an external drive but also syncing files to my laptop and phone too. It supports easy restoring as well.
No traffic ever leaves your local network and since it uses rsync under the hood the devices being sync'd to don't need to run anything other than SSH.
It's a single file shell script that has no dependencies except rsync. It's literally 1,000+ lines of defensive checks and validations to make sure you're not shooting yourself in the foot with rsync, and at the end the last line of code directly calls rsync. It doesn't try to reinvent the wheel by replacing rsync (it's an amazing tool).
Thanks, it's always fun when you're scratching your own itch.
It's also a nice excuse to build in quality of life features that don't take a lot of time because you're using the thing all the time. My favorite one is the color coded rsync command output when DEBUG=1 is set so you can be absolutely sure your config values are producing the expected rsync flags and args.
I reimagined https://searchcode.com/ since I realised LLMs have issues when it comes to understanding code you want to integrate. It’s useful for looking though any codebase, or multiple without having to clone it.
I use it when I have candidate libraries to solve a problem, or I just want to find out how things work. Most recently I pointed it at fzf and was able to pull the insensitive SIMD matching it uses and speed my own projects up.
I can’t find it right now, but there was a post about how ripgrep worked from a someone who walked through the code finding interesting patterns and doing a write up on it. With this I get it over any codebase I find interesting, or can even compare them.
Games. Well, mostly tooling surrounding them it seems. In the last 2 months I've made a pixel art editor for Android, a headless population simulator(still balancing parameters on this one, not enough NPC's turn to crime at present, and I've also run into some weird issues with market prices, in one instance the price of meat rose enough to cause a integer overflow. I could switch to i64, but honestly meat was supposed to cost around 20 moneys, not 2³²
I'm also working on a 2d procedural animation plugin for bevy, a autotiling plugin for bevy (using 16 tile-dual grid, which the default bevy autotiling plug-in didn't support) and ofc my android pixel editor now has a rig editor mode and a tile editor mode that integrates with the plugins.
Making video games is hard! I keep getting side tracked!
I'm not a fan of the TUI form factor for longer running, more ambitious features. Even with a classic "Add an endpoint, tweak the infra, consume in the frontend", plans get awkward to refine in markdown files, especially if everything lives in its own repo.
I wanted something like Plannotator, that could also work for the execution, not just the planning, So I've been working on something that turns Notion into the memory and orchestration layer for agents.
Underneath, it's a plan-implement-review loop, but you get a nice Notion page with a kanban board out of it. You can easily link your existing documentation, collaborate by sharing the page, annotate and comment to steer the planner, and you get versioning out of the box.
Because Notion acts as the memory, you can just open the page after a long weekend and get your agent and yourself back into the full context. You can see what's been done, what's left, or what requires human input just by looking at the board. You can ask it to fetch the comments on the pull request you raised, and it'll fetch, validate the comments, give you a report, and update the plan/board if necessary.
I've been using it exclusively for the last two weeks, I'm quite happy with it. It's been really fun to build the exact tool I wanted.
I’ve got a decent amount of people on the newsletter so trying to figure out how to best deliver indie games via that channel and in the end get more people playing these awesome games people develop :)
Mostly playing around with AI agents session logs.
Lately I’ve been having LLMs implement multiple analysis methods on my session transcripts, trying to surface and identify patterns.
It’s been interesting. It took quite a bit of nudging, but Claude applied techniques I didn’t expect, from disciplines I wouldn’t have thought of.
If it works out, I’d like to turn into a sort of daemon that locally runs analysis on the sessions of users, with a privacy-preserving approach (think federated machine learning).
Would be interesting to see what patterns appear at scale, and have those confirmed or rebutted across thousands of transcripts corpuses. No reason Anthropic & OpenAI should be the only ones to benefit from that; those are our interactions after all.
Working on some improvements to my video platform, https://www.kollaborate.tv . It’s a new video player with side-by-side playback comparison. Claude was really helpful at getting the drift adjustment working because I can push it further than I would be comfortable pushing a human employee in order to get things just right.
In essence, it runs on your mobile device and stores all your data locally. It only connects to the freely available CoinGecko API (for latest prices) and GitHub (for reference and historical data). A background job updates GitHub ref data hourly. There's no login, no cloud, no ads, etc.
I don't have a lot to show for it yet, but I'm working on an online video course for software engineers aspiring to build their own CPU on an FPGA dev board.
Building a map and text-based mobile game where you walk around and graffiti tag things (like Pokemon Go, except you are not looking at a map on your screen). The interface is text room names + descriptions, like an old school MUD, that update as you walk in different directions. They rooms are based partly on what is there in real life, although known points of interest are changed to fit a 'cyberpunk' theme.
The app is built in React Native (almost entirely with AI although I'm fairly particular about some of the features and methods it uses) with a Go backend. Map data comes from PMTiles.
It's a mostly vibe-coded fan site for jamtronica greats Lotus. I wrote/prompted a scraper to pull in setlist data from Nugs and have been having a lot of fun coming up with cool data analysis stuff to do with their sets.
I've seen them 7 times (chump change compared to some fans) and was starting to get certain intuitions about like, "if I hear song X that probably means they won't play song Y." For example, one of my favorite Lotus tunes, It's All Clear To Me Now, seems to fulfill a similar "function" as another song - Did Fatt.
It was pretty cool to see that intuition bear out in the data (they've only ever been played in the same show one time in over 900 total shows).
I've got a bunch of other "data" features sitting in a PR in my Gitlab, need to get around to reviewing and testing it so I can push out the next update. Also have a few other ideas for it, although I think there's probably a point coming fairly soon where there's not really anything left to do.
I posted it on the main Lotus fan group on Facebook. I have a grand total 8 users. I love those users.
The site is nothing crazy, it will never make money or anything - but it's just been a ton of fun to have something cool to hack around on.
Scheduled encrypted back up of git repos via ssh/rsync to a simple server from a macOS workstation. I’m tired of the complexity to host a simple private git repo. Using this suite of scripts, I’ve been able to incrementally backup an encrypted copy of my private git repos to rsync.net (but it could be configured to be any ssh host with rsync capability).
I’ve started moving off WordPress to Yapress. It’s a Git-managed static setup with a migration script, though I haven’t run the full migration yet. Right now, I’m testing the setup and validating the workflow.
The trade-off seems reasonable so far. By going static, the main thing I lose is comments.
The project is still in progress, but I made solid progress over the weekend.
Two things, one is a container control plane inspired in the efforts of the Nextcloud AIO people called LOOM (yeah, like the Lucas Arts game), the other is a full blow NixOS deployment system (from the USB or network directly) for my company so we can deploy the computers for each colleague faster.
On making tools for things I once had to do iteratively. I want to use some of my free time to see if people will use my programs.
The last one I have been working on is a curve-fitting web app (https://fittapp.streamlit.app/).
I do a lot of data science and analytics in my real job.
Download selling tool where you act as your own seller but get tax help and AI support. Much cheaper than the usual suspects and no sales tax for the most part.
Runway, a CRM. It's looking great, near to its first public release, built on market standards. If anyone's interested just ping me, mail handler on profile, ciao.
p.s. still wondering about the licensing to adopt to balance different matters/desires.
It lets you create TV channels from digital media such as YouTube, The Internet Archive, TikTok, Twitch, and Dailymotion. It does that by letting you schedule videos against a custom calendar system.
Since filling out even a month of content can be a lot of work, I built some things to make the process easier.
* Advanced scheduler to know when and how long content can be played at any given datetime
* Real time team collaboration
* Channel libraries to organize media
* "Blocks" - Create a dynamic schedule which generate hours of content that mimics real television scheduling. It even carries over your playback history between generations so that playlists continue from where they left off.
* A catalog to find media from official sources on YouTube
* Embeddable as an OBS browser source to restream your owned content
* Repeat content infinitely or temporarily to create 24/7 channels.
If all goes well I am hoping to re-release sometime this month.
An app for supplementing learning in my masters program [1].
I'm currently in enrolled in the MCS Online from UIUC. My first course, Natural Language Processing, has been interesting, but it's a coursera-based course. This means the lectures are pre-recorded and mostly just the professor reading the slides. It's hard for me to stay engaged and really learn the material. So I started with a series of claude prompts that took the lecture slides and created a pre-watch summary, and then helped me drill the concepts after each lecture. I think converted those into a platform where I can upload notes/lecture slides and have it generate quizzes. It starts with recognition(multiple choice) questions, and eventually moves to recall(short answer) once you prove mastery of a topic. It also generates flashcards from failed answers. It extracts topics from the uploaded materials, and tracks mastery over time. Mastery rots if you don't touch the platform/topic for a while.
I'm not sure if I'll every productize it in any way, but I could see a world where it's used by people prepping for the bar, med boards, various continuing education stuff. Right now it's just a fun platform to build on as I explore the current wave of technologies. Building a framework for evaluating different LLMs for best price/accuracy. Adding a RAG pipeline so wrong answers can point back to source material for further review, etc.
I'm looking at moving from backend engineering to a more MLE or agent pipeline role, so this is giving me something more than school projects to build on. While also helping me do better at school.
I've worked with data my entire career. We need to alt tab so much. What if we put it all on a canvas? Thats what I'm building with Kavla!
Right now working on a CLI that connects a user's local machine to a canvas via websockets. It's open source here: https://github.com/aleda145/kavla-cli
Next steps I want to do more stuff with agents. I have a feeling that the canvas is an awesome interace to see agents working.
VCamper: use LLMs to spot security fixes before CVE publication
Once a patch for a security vulnerability is public, the patch itself can reveal the vulnerability before the CVE is published. VCamper uses a staged LLM pipeline to analyze a Git commit range and flag likely vulnerability patches, even when they look like routine changes.
It’s still a proof of concept, but on known cases like curl CVE-2025-0725 it got close to the published root cause from the patch alone.
This matters because LLMs could make it much harder to keep security fixes quiet: once the patch is public, the bug may be recoverable almost immediately. Quietly shipping a fix and hoping it stays under the radar may stop being a reliable strategy.
I'm building my ideal backend for small projects and hobby stuff. It's inspired by PocketBase, but built around Lua scripting instead of built-in endpoints or usage as a Go library.
Like PocketBase, it's made in Go, has an admin panel, and compiles down to one executable. Here, you write your endpoints as Lua scripts with a simple API for interfacing with requests and the built-in SQLite database. It's minimal and sticks close to being a bare wrapper around the underlying tech (HTTP, SQL, simple file routing), but comes with some niceties too, like automatic backups, a staging server, and a code editor inside the admin panel for quick changes.
It comes from wanting a server that pairs well with htmx (and the backend-first approach in general) that's comfy to use like a CMS. It's not exactly a groundbreaking project, and it still has a ways to go, but I think it's shaping up pretty nicely :)
I’m working on Flowtelic. A workflow driven note-taking system that aims to get you thinking deeper, but also help you work on the most important thing next if you’re stuck. While not essential, it’ll be enhanced with a local first AI approach.
I'm researching Luddite-style examples from around the world. That is, examples of when people rebel against new technology that they see as harming their livelihoods.
Building SiteSecurityScore (https://www.sitesecurityscore.com). A website security scanner that grades your site and tells you exactly what to fix.
It gives you a detailed breakdown of what's missing, step by step guidance on how to fix each issue, and shareable report links. Excellent resource for security teams of all sizes.
Scans HTTP headers, TLS/SSL, DNS security, cookies, and page content. Free to get started, with a REST API for integrating scans into your CI/CD pipeline or monitoring. Also supports capturing and reporting CSP violations.
I built a MacOS-native app [1] to control Positive Grid Spark amps [2], without needing a phone.
Official app is mobile-only and clunky, and the workflow is awkward if you're sitting at a desk. Hardest part has been maintaining compatibility across amp models. Small protocol changes or optimizations I make for one amp can break another. That means I have to do a lot of manual testing before every release. So I'm trying to think of an emulation layer or test harness I can build to make my life easier. Happy to hear suggestions there.
About ~50 people are using it so far, and main feedback has been that it's much faster and more reliable than the official app.
What a cool idea. How does it work? AFAIK The human brain at least does sparse backprop and has SOME neural circuits that feed-backward, so how do you manage it without anything?
Still working https://pagewatch.ai/ , my ai readiness audit tool.
Currently having fun building a MCP app for it inside Claude / ChatGPT, its oddly tricky to get things behaving consistently.
Real challenge to keep it working 24/7. The Android OS, and its modifications are really aggressive, trying to kill everything that runs more than they think it is allowed to.
It's still VERY much in development but I'm building a site that allows people to find TTRPG games that are suited to them AND includes a suite of tools for both GMs and players in said games.
Players will be able to showcase characters they're playing or have played and GMs can manage campaigns (scheduling, notes). I'm a D&D player but I'm trying to make it system-agnostic
Orange Words. My hobby project, a hacker news search system. It was initially created by hand and now I use AI augmented development. It's a good low risk environment for experimenting.
Hey all, I made a free daily ecology cascade game. https://Trophle.com
I'm in school for environmental science and I want to make educational games and resources
I work in robotics and with quaternions (mainly 6DoF SLAM and used to do robot arm kinematics), but I don't get the use case for this. Maybe provide some example use cases?
Six months ago, that would have been unrealistic, because we're heavily committed to the mongodb API and we make it part of our own API.
Starting in December though, Opus 4.6 made it perfectly realistic to pursue this with Claude Code as a series of personal weekend projects.
Now, despite not having any official resources on this until the last week or so, it should land in May.
This doesn't work for everything. It absolutely helps that the problem I'm solving is an "adapter pattern" problem: "make X talk like Y." And that we have a massive test suite, at multiple levels. That combination makes "here's the problem, go solve it, grind until the tests pass, don't bother me for a few hours" a realistic AI agent request.
But it's a little mind-blowing all the same. The hype around AI is so out of control, it can be easy to miss genuine "holy crap" moments.
Along the way I've written a fair bit about how to run Claude Code autonomously on your household server in a reasonably secure manner:
A software for managing BJJ and martial arts academies that it's both easy to use and have everything they need like assistance tracking, payments, communications, etc.
It's called MatGoat[1], and it's going quite well so far. Nowadays I'm working more on the marketing/sales side.
Swiss army knife CLI tool written in Swift using only native Apple frameworks.
The primary goal of this project is to demonstrate how many Apple standard library frameworks can be meaningfully used in a single, actually-useful CLI tool.
I'm building a debate/writing game platform: https:argyu.fun
The mission is to incentivize better thinking. For each game there's an AI judge that scores everyone's answer based on a public rubric (style, cohesion, logic, etc).
Currently uses fake money and ELO score but thought it could be a very interesting competitive game for real stakes.
Building something that finally stops making me the tester for my own AI. You know that moment where the AI finishes writing code and then goes "can you run this and check if it works?" I got tired of that loop. So I built an IDE that just... runs it, clicks through it, finds what broke, and fixes it. You watch.
Not Better Cursor , But what comes after it.
I've got this new account and a Substack page where I'm writing about, idk... metaphysical stuff? Spirituality, religion, psychedelics, tarot, and so forth. I was inspired largely by the Weird Studies podcast, but there's a bunch of actually interesting writing and media in this space right now.
I deliberately separated it from my public internet persona (which is connected to my real name) in the hopes that I could write about weird, woo-y, or controversial topics without worry. I've got a few articles half baked and have been having fun engaging with a different subset of the Substack crowd than my normal tech focus would show me.
Of course the stats show that the one article I did that touches on AI has done an order of magnitude better than anything else.
Anyway this is just kind of a weird sideline project, a sort of release valve for stuff that wouldn't fit in on my "professional" site, but it's been a fun thing to spend some time on.
Another thing that's cool is that I largely stopped _writing_ a few years back. I always enjoyed writing but of course as a dev most of my stuff had a technical/tutorial bent to it. Writing weird little "what do I think" essays has forced me to exercise a writing muscle I really hadn't stretched for a long time and I've enjoyed it.
There's only a handful of things up now, it's nothing special really. Link in my bio, if you see something you like I would love to hear from you!
From another submission (https://news.ycombinator.com/item?id=47738827), there was a screenshot of Google Docs/Drive showing a popup saying "You cannot do copy/cut/paste with the mouse" whenever you try to right-click and copy.
Some months ago, I saw that very popup, and finally started working on something I've been wanted to do for a long-time, a spreadsheet application. It's cross-platform (looks and work identical across Windows, macOS and Linux), lightweight, and does what a spreadsheet application should be able to do, in the way you expect it, forever. As an extra benefit, I can finally open some spreadsheets that grown out of control (+100MB and growing) without having to go and make a cup of coffee while the spreadsheet loads.
I don't really have any concrete to share, I guess it'll be a Show HN eventually, but I thought it was funny it was brought up in a similar way in that article as was the motivation for me to build yet another spreadsheet application.
It's an LLM-webapp-builder, sure, but different from the rest! I have the LLM write python code when it needs to modify an HTML file for example (it'll use beautifulsoup; then I run the code: it parses the source into a data structure, modifies the data structure, and then outputs the resulting html).
It's also a marketplace where you can publish your llm-powered webapp, and earn $ on the token margins (I charge 2x token rates) when people use your site.
Building a pro transparency writing tool that cryptographically proves a human actually typed what they claim to have written (research papers, news articles, assignments etc) . It captures behavioral signals during composition, makes it very hard to automate or fake the writing process, and lets readers verify authorship authenticity. Think "proof of human work" for the AI generated slop era.
Codify — democratic digital public infrastructure that turns your problems into structured, executable programs.
The idea: describe any problem in plain language (voice or text), and AI codifies it into a structured program with the right people, steps, timeline, and agents to get it done. It's a 5-step wizard: Define Problem → Codify Solution → Setup Program → Execute Program → Verify Outcome.
It runs across 50+ domains — codify.healthcare (EMR backend), codify.education (LMS backend), codify.finance, codify.careers (HRM backend), codify.law, plus 13 city domains (codify.nyc, codify.miami, codify.london, codify.tokyo, etc.). Each domain tailors the AI assessment and program output to that sector.
The platform is Project20x — think of it as the infrastructure layer. If Codify is the verb ("codify your healthcare problem into a care program"), Project20x is the operating system that runs it all: multi-tenant governance, AI agent orchestration, and domain-specific sys-cores for healthcare, education, city services, etc.
Every US federal agency and state-level department has a subdomain — ed.usa.project20x.com (Dept of Education), doj.usa.project20x.com, hhs.usa.project20x.com, etc. — with AI agents representing each agency's mandate. Same structure at the state level.
The political side: Project20x hosts policy management for both parties — dnc.project20x.com and rnc.project20x.com — where legislative intent gets codified into executable governance through a 10-step policy lifecycle. Right now I'm building out the multi-agent environment so agency agents can negotiate with each other, make deals, and send policy proposals up to the HITL (human-in-the-loop) politician for approval. Each elected official has a profile (e.g. https://project20x.com/u/donald-trump) where constituents can engage and where policy proposals land for review.
The name is a nod to structured policy frameworks, but the goal is nonpartisan infrastructure: democratically governed essential services delivered as AI-native social programs.
Stack: Nuxt 2/Vue 2 frontend, Laravel 10 API, Python/LangGraph agent orchestration, Flutter mobile app. Currently live across all domains.
I wanted to make JSON/YAML configuration language for my projects. And i wanted a strict specification. This is want i created, now with specification and 100% coverage, reference implementation it’s just one prompt to reimplement parser in another language.
I wanted to make it easier to quickly see/study trending articles on Wikipedia because they tend to make good topics to know before going to trivia night.
I've had the domain for awhile, but just made the app today on a whim.
I use Wikimedia's api to get the trending articles, curate them a bit, add some annotations to provide some context, then push to deploy the static site.
https://tiledwords.com
After winning the Playlin Player's Choice award I've noticed an uptick in players as well as some people sharing videos on YouTube which has been fun. I've got a few thousand people playing every day.
I just launched user accounts today so user's can now track their progress across devices and share their stats with each other. This ended up being a bigger chunk of work than I expected but I'm really pleased with how it turned out. (Though I launched it 15 minutes ago so I'm holding my breath for bug reports)
I'm fine-tuning my internal puzzle-building now with the goal of letting people use them to make and share their own puzzles soon!
It's an iOS app that applies various generative art effects to your photos, letting you turn your photos into creative animated works of art. It's fully offline, no AI, no subscriptions, no ads, etc.
I'm really proud of it and if you've been in the generative art space for a while you'll instantly recognise many of the techniques I use (circle packing, line walkers, mosaic grid patterns, marching squares, voronoi tessellation, glitch art, string art, perlin flow fields, etc.) pretty much directly inspired by various Coding Train videos.
Direct download link on the App Store is https://apps.apple.com/us/app/photogenesis-photo-art/id67597... if you want to try it out.
* Coming to Android soon too.
Since your app is fully offline I'd love to chat about photogenesis/your general work in this area since there may be a good opportunity for collaboration. I've been working on some image stuff and want to build a local desktop/web application, here are some UI mockups of that I've been playing with (many AI generated though some of the features are functional, I realized that with CSS/SVG masks you can do a ton more than you'd expect): https://i.imgur.com/SFOX4wB.png https://i.imgur.com/sPKRRTx.png but we don't have all the ui/vision expertise we'd need to take them to completion most likely.
When booking flights, I use sites like Kiwi and Skyscanner that let you do flexible searches - multiple destinations, custom connections, creative routes, etc. But rail search feels oddly constrained. All the UK train operators offer basically the same experience, and surface the exact same routes. I always suspected there were better or just different options that weren’t being shown. Where is the "Skyscanner for trains"?
After digging through the national rail data feeds, I decided to have a go at building my own route planner that runs completely offline in the browser. This gave me the freedom to implement more complex filters, search to/from multiple stations, and do it without a persistent network connection.
Now I'm finding routes that aren't offered by the standard train operators, connecting at different stations, and finding it's often easier to travel to different stations (some I'd never heard of) that get me closer and faster to where I actually want to go!
It's still a little rough and I'd like to add more features such as fares, VSTP data, and direct-links to book tickets, but wanted to share early and get some initial feedback before investing more time into it. So, thanks in advance - let me know what you think.
https://playruly.com
My goal is to make a simple yet interesting procedural and replayable puzzle. It has a couple of weekly variations: on Saturdays you need to break a rule to score max points, and on Mondays there's an added memory aspect which brings variety to the game.
It's mostly vibe-coded which lets me focus on game design and testing. The next step is better onboarding/tutorial and more intuitive UI.
It scans your claude and codex history to find edits and matches those to git commits (even if the code was auto-formatted).
https://buildermark.dev/
You can browse all 364 prompts that wrote 94% of the code here:
https://demo.buildermark.dev/projects/u020uhEFtuWwPei6z6nbN
https://github.com/jank-lang/jank
It's a native Clojure dialect which is also a C++ dialect, including a JIT compiler and nREPL server. I'm currently building out a custom IR so I can do optimization passes at the level of Clojure semantics, since LLVM will not be able to do them at the LLVM IR level.
You can read more about it over at the site, but it allows you to construct and validate arguments in a graphical form, and it has truth/proof propagation so you can see whether a conclusion is currently considered valid or contested. Some upcoming plans are to allow users to validate arguments for themselves, like mark which parts they understand and agree with so they can collapse that part of the graph, and to add more mcp capability so that LLM can help you construct and validate new arguments.
https://github.com/BVCampos/operator
It has been working quite well.
I had an insight the other day, that as I fix the n least (and most, it's a palindrome!) significant decimal digits, I also fix the remainder from division in 5^n. Let's call it R. Since I also fix by that point a bunch of least (and most) significant bits, I can subtract how much they contribute mod 5^n from R, to get the remainder from division in 5^n of the still unknown bit. The thing is, maybe it's not possible to get this specific remainder with the unknown bits, because they're too few.
So, I can prepare in advance a table of size 5^n (for one or more ns) which tells me how many bits from the middle of the palindrome I need, to get a remainder of <index mod 5^n>.
Then when I get to the aforementioned situation, all I need to do is to compare the number in the table to number of unknown bits. If the number in the table is bigger, I can prune the entire subtree.
From a little bit of testing, this seems to work, and it seems to complement my current lookup tables and not prune the same branches. It won't make a huge difference, but every little bit helps.
The important thing, though, is that I'm just happy there are still algorithmic improvements! For a long while I've been only doing engineering improvements such as more efficient tables and porting to CUDA, but since the problem is exponential, real breakthroughs have to come from a better algorithm, and I almost gave up on finding one.
[0] https://ashdnazg.github.io/articles/22/Finding-Really-Big-Pa...
https://ewams.net/?date=2026/03/29&view=Qwen35_Performance_w...
Imagine mixing Magic: The Gathering, StarCraft and Civilization’s hex grid combat.
There’s multiplayer but I haven’t put the server anywhere yet.
Check out the introduction here:
https://github.com/williamcotton/space-trader/blob/main/docs...
Clone the repo:
There’s maybe a couple of other games called Space Trader so if anyone has any suggestions for a new name, I’m all ears!Model output volumes mean that code review only as a final check before merge is way too late, and far too burdensome. Using AI to review AI-generated code is a band-aid, but not a cure.
That's why I built Caliper (http://getcaliper.dev). It's a system that institutes multiple layers of code quality checks throughout the dev cycle. The lightest-weight checks get executed after every agent turn, and then increasingly more complex checks get run pre-commit and pre-merge.
Early users love it, and the data demonstrates the need - 40% of agent turns produce code that violates a project's own conventions (as defined in CLAUDE.md). Caliper catches those violations immediately and gets the model to make corrections before small issues become costly to unwind.
Still very early, and all feedback is welcome! http://getcaliper.dev
It has some interesting applications for building high performance clients for mssql with tds protocol implementation. The APIs allow almost direct data serialization to wire instead of datatype materialization in rust. Makes for a suitable contender for high performance language interop.
I want to show how I liberate poorly aligned, pixelated PDF image scans of century-old Latin textbooks from the Internet Archive and transform them into glorious Org mode documents while preserving important typographic details, nicely formatted tables, and some semantic document metadata. I also want to demonstrate how I use a high-performance XML database engine to quickly perform Latin-to-English lookups against an XML-TEI formatted edition of the 19th century Lewis & Short dictionary, and using a RESTXQ endpoint and some XQuery code to dynamically reformat the entries into Org-mode for display in a pop-up buffer.
I intend demonstrate how I built a transcription pipeline in Emacs Lisp using tools such as yt-dlp and patreon-dl to grab Latin-language audio content from the Internet, transcode the audio with ffmpeg, do Voice Activity Detection and chunking in Python with Silero, load the chunks into Gemini's context window, and send it off for transcription and macronization, gather forced-alignment data using local a local wav2vec2-latin model, and finally add word-level linguistic analysis (POS, morphology, lemmas) using a local Stanza model trained on the Classical corpus.
This all gets saved to an an XML file which is loaded into BaseX along with some metadata. I'll then demonstrate some Emacs Lisp code which pulls it into an Org-mode based transcription buffer and minor-mode for reading and study, where I can play audio of any given Latin word, sentence, or paragraph, thanks to the forced-alignment and linguistic analysis data being stored in hidden text properties when the data was fetched from the database.
Lastly, I'd like to explore how to leverage these tools to automatically create flash cards with audio cues in Org mode using the anki-editor Emacs minor mode for sentence mining.
Here's the MVP interface: https://bcmullins.github.io/reading/
I appreciate any feedback. Hope you find something interesting to read!
Also, Arch Ascent, which is a tool for evolveing microservice-heavy architectures.
https://github.com/mikko-ahonen/arch-ascent/blob/main/doc/de...
https://github.com/flipbit03/terminal-use
I'm super proud, because it came to my knowledge that someone at Codex used my tool to debug codex+zellij issues, by running zellij within `tu`, and then codex inside zellij
I just published a fun interactive 3D demo of SPDC, one of the most common and accessible ways to create entangled pairs of photons. I'm hoping to publish a series of articles on other cool learnings about doing quantum photonics in the lab.
https://paulg.info/2026/04/10/spdc/
https://youtu.be/hYyrgDEJLOA
The use case is kind of neat. RAID is great and can mostly solve these problems, but people don't have SATA hardware that can handle the workload well, plus they aren't ready to manage an array like that, and they don't like having to use specific sized drives, etc. Another major issue with those setups is you need to be careful because an IO error that you don't recover from will be very difficult or impossible to recover from because of the layers of LUKS combined with LVM.
With MergerFS you just use regular file systems that are separate, but they get combined into a single mount point. That means each disk can just be a different LUKS encrypted drive and if you need to recover data it's isolated to that one disk and much more manageable. You can also take any disk and plug it into another machine as needed and grow or shrink the storage pool as needed.
MergerFS has options and settings to help you determine how files are spread across the drives, such as least space used or which disk has the most of that directory path already.
My app (Chimera) automates the unlocking of the disks, mounting and some data migration if you want to remove a disk from the pool. I plan to add some rclone features to help provide easier backup options to places like Backblaze, AWS, or a remote server in general.
So far so good and I was surprised at how well Opus had been handling Gtk and pkexec.
Let me know if you guys are interested I am close to pushing some RPMs and DEBs, in addition to the standard Python stuff.
I think it works quite well so far, but need to tweak the camera algorithm a bit to make the buttons work better. Thinking about more games to add as well.
3 days ago, 220 comments: https://news.ycombinator.com/item?id=47700460
5 days ago, 51 comments: https://news.ycombinator.com/item?id=47679021
8 days ago, 21 comments: https://news.ycombinator.com/item?id=47639039
11 days ago, 22 comments: https://news.ycombinator.com/item?id=47600204
It allows you to get a wake up call from someone friendly, somewhere out there in the world.
It's got a handful of regular users and it's mostly me making the calls, but it's great fun to wake people up!
No phone number required - these are VoIP calls via the app.
Built it because I think it's cool.
https://www.roastrack.app/
It's a free USCIS form-filling web-app(no Adobe required). USCIS forms still use XFA PDFs, which don’t let you edit in most browsers. Even with Adobe, fields break, and getting the signature is hard.
So I converted the PDF form into modern, browser-friendly web forms - and kept every field 1:1 with the original. You fill the form, submit it, and get the official USCIS PDF filled.
I found out SimpleCitizen(YC S16) offers a DIY plan for $529 [2]
So, a free (and local-only) version might be a good alternative
[1] https://fillvisa.com/demo
[2] https://www.simplecitizen.com/pricing/
I use it daily and so do others, for - better UX, feedback, and review surfaces for ai coding agents.
Free and open source https://github.com/backnotprop/plannotatorUnlike those apps it has full support for design tokens and (so far) flexbox layouts. It can also export directly to HTML, rather than a fake preview mode. I’m also working on full code-backed components, so you can go between code and design very easily.
As a designer, I’ve been frustrated for years by the gap between design and code, and despite all the new AI features, Figma still hasn’t got any further in years - design tokens need a 3rd party plugin and responsive designs are a pain in the bum. So I decided to build something that has the ease of Figma while being much closer to live code.
I’ve got to the point where I’m designing the app in itself, tokens are working, html export is working and nearly ready for first betas.
PSA PS. Don't post generated/AI-edited comments. HN is for conversation between humans https://news.ycomtem?id=47340079
https://GetSetReply.com/
I am hoping to launch in about a week, so I would love any user feedback! (email in profile)
Code: https://github.com/opensciencearchive/server.
Website: https://opensciencearchive.org/
Two demos:
https://pockets.bio
https://lingual.bio
I've got demos up and running (mirroring/extending PDB and GEO). Next I'm working on APIs with good AX, ML-friendly export, and an unified AI-driven UI that works for all scientific data types.
Already using it for my SQLite driver, and already in use by some a few other projects: https://github.com/topics/wasm2go
Right now I'm focused on the stats side. It already shows how much time you spend in each app, and I'm adding website tracking too, which should make the picture much more useful.
I'm also working on better break timing for dictation. LookAway already delays a due break if you're in the middle of typing, so it does not interrupt at a bad time. Now I'm trying to extend that same behavior to dictation as well, which turns out to be a pretty interesting detection problem because it overlaps with some of the other context signals I already use.
Most of the challenge is making it smarter without making it feel more intrusive.
https://virissimo.info/build-your-own-alu/
LMK what you think.
https://www.metanoia-research.com/
It includes bill of materials, purchase/production orders, "can I make n?", stock takes, multiple stock locations, and barcode scanning. It's aimed mainly at small business and makers for the time-being, but still allows multiple users to connect over the the local network.
Eventually I got scope creeped into a full game with branching stories, item crafting, and a custom cutscene engine...even Trained a model for a few specific art assets.
https://store.steampowered.com/app/4301600/Cherrys_Dungeon/
https://github.com/Lramseyer/vaporview
Repo: https://github.com/jbonatakis/pginbox
Makes reading/searching the Postgres mailing lists easier.
I’m polling a Fastmail inbox to nearly instantly receive and ingest messages. Anyone can browse without an account, but registered users can follow threads to be notified of new messages, threads in which your registered email is found are auto-followed, and there are some QOL settings.
Search is pretty naive right now (keyword on subjects) but improved search is the next big thing on my list.
https://prettygoodmusic.app
A work in progress.
One thing I find especially intriguing is how LLMs can help deal with desinformation:
- I experiment with deterministic settings of local LLMs for the document summary so that sharing a prompt would prove that the output was not tempered with (no desinformation on the service side)
- I add outputs of several LLMs (from the US, the EU and from China) for the "broader context" section so users could compare the output (no desiformation on the provider and model side)
Looking for people who know hardware well. Let's get to know one another on a flight to Shenzhen :P
The business model is likely going to revolve around mcp and x402 https://micro.mu/developers/
No traffic ever leaves your local network and since it uses rsync under the hood the devices being sync'd to don't need to run anything other than SSH.
It's a single file shell script that has no dependencies except rsync. It's literally 1,000+ lines of defensive checks and validations to make sure you're not shooting yourself in the foot with rsync, and at the end the last line of code directly calls rsync. It doesn't try to reinvent the wheel by replacing rsync (it's an amazing tool).
It's also a nice excuse to build in quality of life features that don't take a lot of time because you're using the thing all the time. My favorite one is the color coded rsync command output when DEBUG=1 is set so you can be absolutely sure your config values are producing the expected rsync flags and args.
I use it when I have candidate libraries to solve a problem, or I just want to find out how things work. Most recently I pointed it at fzf and was able to pull the insensitive SIMD matching it uses and speed my own projects up.
I can’t find it right now, but there was a post about how ripgrep worked from a someone who walked through the code finding interesting patterns and doing a write up on it. With this I get it over any codebase I find interesting, or can even compare them.
I'm also working on a 2d procedural animation plugin for bevy, a autotiling plugin for bevy (using 16 tile-dual grid, which the default bevy autotiling plug-in didn't support) and ofc my android pixel editor now has a rig editor mode and a tile editor mode that integrates with the plugins.
Making video games is hard! I keep getting side tracked!
I'm not a fan of the TUI form factor for longer running, more ambitious features. Even with a classic "Add an endpoint, tweak the infra, consume in the frontend", plans get awkward to refine in markdown files, especially if everything lives in its own repo.
I wanted something like Plannotator, that could also work for the execution, not just the planning, So I've been working on something that turns Notion into the memory and orchestration layer for agents.
Underneath, it's a plan-implement-review loop, but you get a nice Notion page with a kanban board out of it. You can easily link your existing documentation, collaborate by sharing the page, annotate and comment to steer the planner, and you get versioning out of the box.
Because Notion acts as the memory, you can just open the page after a long weekend and get your agent and yourself back into the full context. You can see what's been done, what's left, or what requires human input just by looking at the board. You can ask it to fetch the comments on the pull request you raised, and it'll fetch, validate the comments, give you a report, and update the plan/board if necessary.
I've been using it exclusively for the last two weeks, I'm quite happy with it. It's been really fun to build the exact tool I wanted.
I’ve got a decent amount of people on the newsletter so trying to figure out how to best deliver indie games via that channel and in the end get more people playing these awesome games people develop :)
Lately I’ve been having LLMs implement multiple analysis methods on my session transcripts, trying to surface and identify patterns.
It’s been interesting. It took quite a bit of nudging, but Claude applied techniques I didn’t expect, from disciplines I wouldn’t have thought of.
If it works out, I’d like to turn into a sort of daemon that locally runs analysis on the sessions of users, with a privacy-preserving approach (think federated machine learning).
Would be interesting to see what patterns appear at scale, and have those confirmed or rebutted across thousands of transcripts corpuses. No reason Anthropic & OpenAI should be the only ones to benefit from that; those are our interactions after all.
Do you have any example?
The query engine itself is like a DAG of 'operators', similar to a relational DB (or more like a graph one) with scanners, filters, and matchers.
Very fun, although not at all efficient and probably overengineered for what it does :)
https://www.geosystemsdev.com/products/hodlings/
In essence, it runs on your mobile device and stores all your data locally. It only connects to the freely available CoinGecko API (for latest prices) and GitHub (for reference and historical data). A background job updates GitHub ref data hourly. There's no login, no cloud, no ads, etc.
The app is built in React Native (almost entirely with AI although I'm fairly particular about some of the features and methods it uses) with a Go backend. Map data comes from PMTiles.
https://lotuseater.epiccoleman.com/
It's a mostly vibe-coded fan site for jamtronica greats Lotus. I wrote/prompted a scraper to pull in setlist data from Nugs and have been having a lot of fun coming up with cool data analysis stuff to do with their sets.
I've seen them 7 times (chump change compared to some fans) and was starting to get certain intuitions about like, "if I hear song X that probably means they won't play song Y." For example, one of my favorite Lotus tunes, It's All Clear To Me Now, seems to fulfill a similar "function" as another song - Did Fatt.
It was pretty cool to see that intuition bear out in the data (they've only ever been played in the same show one time in over 900 total shows).
I've got a bunch of other "data" features sitting in a PR in my Gitlab, need to get around to reviewing and testing it so I can push out the next update. Also have a few other ideas for it, although I think there's probably a point coming fairly soon where there's not really anything left to do.
I posted it on the main Lotus fan group on Facebook. I have a grand total 8 users. I love those users.
The site is nothing crazy, it will never make money or anything - but it's just been a ton of fun to have something cool to hack around on.
Very early demo with a smart dum-dum RL agent here:
https://rlplays.com
The trade-off seems reasonable so far. By going static, the main thing I lose is comments.
The project is still in progress, but I made solid progress over the weekend.
The project is here: https://github.com/yusufaytas/yapress
https://radiantskin.app/
I do a lot of data science and analytics in my real job.
It lets you create TV channels from digital media such as YouTube, The Internet Archive, TikTok, Twitch, and Dailymotion. It does that by letting you schedule videos against a custom calendar system.
Since filling out even a month of content can be a lot of work, I built some things to make the process easier.
* Advanced scheduler to know when and how long content can be played at any given datetime
* Real time team collaboration
* Channel libraries to organize media
* "Blocks" - Create a dynamic schedule which generate hours of content that mimics real television scheduling. It even carries over your playback history between generations so that playlists continue from where they left off.
* A catalog to find media from official sources on YouTube
* Embeddable as an OBS browser source to restream your owned content
* Repeat content infinitely or temporarily to create 24/7 channels.
If all goes well I am hoping to re-release sometime this month.
I'm not sure if I'll every productize it in any way, but I could see a world where it's used by people prepping for the bar, med boards, various continuing education stuff. Right now it's just a fun platform to build on as I explore the current wave of technologies. Building a framework for evaluating different LLMs for best price/accuracy. Adding a RAG pipeline so wrong answers can point back to source material for further review, etc.
I'm looking at moving from backend engineering to a more MLE or agent pipeline role, so this is giving me something more than school projects to build on. While also helping me do better at school.
[1] https://studyengine.app/
Fully local, hobbyist friendly, agentic workflows work great with it since it’s just a CLI.
I've worked with data my entire career. We need to alt tab so much. What if we put it all on a canvas? Thats what I'm building with Kavla!
Right now working on a CLI that connects a user's local machine to a canvas via websockets. It's open source here: https://github.com/aleda145/kavla-cli
Next steps I want to do more stuff with agents. I have a feeling that the canvas is an awesome interace to see agents working.
Built with tldraw, duckdb and cloudflare
https://inSolitaire.com
I am currently rewriting+testing the engine and about to add ~400 games to my platform in a few weeks.
Once a patch for a security vulnerability is public, the patch itself can reveal the vulnerability before the CVE is published. VCamper uses a staged LLM pipeline to analyze a Git commit range and flag likely vulnerability patches, even when they look like routine changes.
It’s still a proof of concept, but on known cases like curl CVE-2025-0725 it got close to the published root cause from the patch alone.
This matters because LLMs could make it much harder to keep security fixes quiet: once the patch is public, the bug may be recoverable almost immediately. Quietly shipping a fix and hoping it stays under the radar may stop being a reliable strategy.
https://github.com/rndhouse/vcamper
Like PocketBase, it's made in Go, has an admin panel, and compiles down to one executable. Here, you write your endpoints as Lua scripts with a simple API for interfacing with requests and the built-in SQLite database. It's minimal and sticks close to being a bare wrapper around the underlying tech (HTTP, SQL, simple file routing), but comes with some niceties too, like automatic backups, a staging server, and a code editor inside the admin panel for quick changes.
It comes from wanting a server that pairs well with htmx (and the backend-first approach in general) that's comfy to use like a CMS. It's not exactly a groundbreaking project, and it still has a ways to go, but I think it's shaping up pretty nicely :)
link: https://github.com/ksymph/mogo
https://www.flowtelic.com
I’m interested too, but don’t have amazing patience to dig into it.
For me this is an example of when you become aware of something you see it all around.
I'll writeup a fuller list and what I learned along the way.
It gives you a detailed breakdown of what's missing, step by step guidance on how to fix each issue, and shareable report links. Excellent resource for security teams of all sizes.
Scans HTTP headers, TLS/SSL, DNS security, cookies, and page content. Free to get started, with a REST API for integrating scans into your CI/CD pipeline or monitoring. Also supports capturing and reporting CSP violations.
Official app is mobile-only and clunky, and the workflow is awkward if you're sitting at a desk. Hardest part has been maintaining compatibility across amp models. Small protocol changes or optimizations I make for one amp can break another. That means I have to do a lot of manual testing before every release. So I'm trying to think of an emulation layer or test harness I can build to make my life easier. Happy to hear suggestions there.
About ~50 people are using it so far, and main feedback has been that it's much faster and more reliable than the official app.
[1] https://tonepilot.app [2] https://www.positivegrid.com/products/spark-2
https://github.com/Findeton/hebbi
I tinkered for a minute but never got anywhere.
Real challenge to keep it working 24/7. The Android OS, and its modifications are really aggressive, trying to kill everything that runs more than they think it is allowed to.
I made a whole article about it. I hope it will help others: https://dev.to/stoyan_minchev/i-spent-several-months-buildin...
Program your amateur radio via the web. Uses pyiodide + chirp drivers under the hood + WebSerial.
It's still VERY much in development but I'm building a site that allows people to find TTRPG games that are suited to them AND includes a suite of tools for both GMs and players in said games.
Players will be able to showcase characters they're playing or have played and GMs can manage campaigns (scheduling, notes). I'm a D&D player but I'm trying to make it system-agnostic
Orange Words. My hobby project, a hacker news search system. It was initially created by hand and now I use AI augmented development. It's a good low risk environment for experimenting.
https://agjmills.github.io/trove/
Go, docker, bit of alpine js
I'm working to make it better right now.
https://channelsurfer.tv/
https://github.com/VoxleOne/SpinStep
https://apostrophecms.com
Six months ago, that would have been unrealistic, because we're heavily committed to the mongodb API and we make it part of our own API.
Starting in December though, Opus 4.6 made it perfectly realistic to pursue this with Claude Code as a series of personal weekend projects.
Now, despite not having any official resources on this until the last week or so, it should land in May.
This doesn't work for everything. It absolutely helps that the problem I'm solving is an "adapter pattern" problem: "make X talk like Y." And that we have a massive test suite, at multiple levels. That combination makes "here's the problem, go solve it, grind until the tests pass, don't bother me for a few hours" a realistic AI agent request.
But it's a little mind-blowing all the same. The hype around AI is so out of control, it can be easy to miss genuine "holy crap" moments.
Along the way I've written a fair bit about how to run Claude Code autonomously on your household server in a reasonably secure manner:
https://apostrophecms.com/blog/how-to-be-more-productive-wit...)
Also general Claude Code tips and thoughts on workflows that help and workflows that ultimately just speed your burnout:
https://apostrophecms.com/blog/claude-code-part-2-making-the...
I know, everybody's writing this stuff, but the desire to share is natural.
(Disclaimer: I'm part of the demographic AI was trained on. If I tried not to sound like a bot, I'd have to sound like... well, somebody else)
It's called MatGoat[1], and it's going quite well so far. Nowadays I'm working more on the marketing/sales side.
[1] https://matgoat.com/en/
Swiss army knife CLI tool written in Swift using only native Apple frameworks.
The primary goal of this project is to demonstrate how many Apple standard library frameworks can be meaningfully used in a single, actually-useful CLI tool.
brew install jftuga/tap/swiftswiss
It's in rust with egui, and should help folks to do that without the cli.
Not ready for prime time yet, but available at https://github.com/almet/signal-without-smartphone
and a gift for my friend's birthday.
- make it reliable to run LLM inference on company hardware, even when it is poor or outdated
- bring chaotic agentic behavior under control in business contexts
https://github.com/netdur/hugind
The mission is to incentivize better thinking. For each game there's an AI judge that scores everyone's answer based on a public rubric (style, cohesion, logic, etc).
Currently uses fake money and ELO score but thought it could be a very interesting competitive game for real stakes.
Any feedback is much appreciated.
https://buttplug.io
https://cybernetic.dev/cube
Posted a show hn earlier today that didn't got any traction : https://news.ycombinator.com/item?id=47738516
https://fluenly.ai/
https://grge.github.io/weierstrass/
Have you heard of Superformula ? I remember playing with them few years ago.
Turns your project's GitHub release notes into user changelog that your users actually want to read.
A tool to estimate if you should vibe an automation/app or just buy/delegate/grind instead
I deliberately separated it from my public internet persona (which is connected to my real name) in the hopes that I could write about weird, woo-y, or controversial topics without worry. I've got a few articles half baked and have been having fun engaging with a different subset of the Substack crowd than my normal tech focus would show me.
Of course the stats show that the one article I did that touches on AI has done an order of magnitude better than anything else.
Anyway this is just kind of a weird sideline project, a sort of release valve for stuff that wouldn't fit in on my "professional" site, but it's been a fun thing to spend some time on.
Another thing that's cool is that I largely stopped _writing_ a few years back. I always enjoyed writing but of course as a dev most of my stuff had a technical/tutorial bent to it. Writing weird little "what do I think" essays has forced me to exercise a writing muscle I really hadn't stretched for a long time and I've enjoyed it.
There's only a handful of things up now, it's nothing special really. Link in my bio, if you see something you like I would love to hear from you!
Deployment tool with security gates.
Some months ago, I saw that very popup, and finally started working on something I've been wanted to do for a long-time, a spreadsheet application. It's cross-platform (looks and work identical across Windows, macOS and Linux), lightweight, and does what a spreadsheet application should be able to do, in the way you expect it, forever. As an extra benefit, I can finally open some spreadsheets that grown out of control (+100MB and growing) without having to go and make a cup of coffee while the spreadsheet loads.
I don't really have any concrete to share, I guess it'll be a Show HN eventually, but I thought it was funny it was brought up in a similar way in that article as was the motivation for me to build yet another spreadsheet application.
It's an LLM-webapp-builder, sure, but different from the rest! I have the LLM write python code when it needs to modify an HTML file for example (it'll use beautifulsoup; then I run the code: it parses the source into a data structure, modifies the data structure, and then outputs the resulting html).
It's also a marketplace where you can publish your llm-powered webapp, and earn $ on the token margins (I charge 2x token rates) when people use your site.
The idea: describe any problem in plain language (voice or text), and AI codifies it into a structured program with the right people, steps, timeline, and agents to get it done. It's a 5-step wizard: Define Problem → Codify Solution → Setup Program → Execute Program → Verify Outcome.
It runs across 50+ domains — codify.healthcare (EMR backend), codify.education (LMS backend), codify.finance, codify.careers (HRM backend), codify.law, plus 13 city domains (codify.nyc, codify.miami, codify.london, codify.tokyo, etc.). Each domain tailors the AI assessment and program output to that sector.
The platform is Project20x — think of it as the infrastructure layer. If Codify is the verb ("codify your healthcare problem into a care program"), Project20x is the operating system that runs it all: multi-tenant governance, AI agent orchestration, and domain-specific sys-cores for healthcare, education, city services, etc.
Every US federal agency and state-level department has a subdomain — ed.usa.project20x.com (Dept of Education), doj.usa.project20x.com, hhs.usa.project20x.com, etc. — with AI agents representing each agency's mandate. Same structure at the state level.
The political side: Project20x hosts policy management for both parties — dnc.project20x.com and rnc.project20x.com — where legislative intent gets codified into executable governance through a 10-step policy lifecycle. Right now I'm building out the multi-agent environment so agency agents can negotiate with each other, make deals, and send policy proposals up to the HITL (human-in-the-loop) politician for approval. Each elected official has a profile (e.g. https://project20x.com/u/donald-trump) where constituents can engage and where policy proposals land for review.
The name is a nod to structured policy frameworks, but the goal is nonpartisan infrastructure: democratically governed essential services delivered as AI-native social programs.
Stack: Nuxt 2/Vue 2 frontend, Laravel 10 API, Python/LangGraph agent orchestration, Flutter mobile app. Currently live across all domains.
https://project20x.com | https://codify.healthcare | https://codify.education | https://dnc.project20x.com | https://rnc.project20x.com etc...
https://drivelens.click/
No file contents are accessed, only metadata, fully client-side API calls (browser to google API).