Ask HN: What are you working on? (May 2026)

Posted by david927 18 hours ago

What are you working on? Any new ideas that you're thinking about?

190 points | 705 commentspage 2

hy-token 1 hour ago|

I am working on Liel (https://hy-token.github.io/liel/), a tool that can compare and integrate LLM memory as single file.

Instead of saving LLM memory in Markdown, I want to manage it using a graph structure to easily record the relationships between tasks and decisions, and persist when, why, and how they changed.

Asmod4n 1 hour ago||

I'm building a small framework to make desktop apps in mruby, targeting Linux, Mac and Windows. The aim is a native app for the kind of thing you'd usually ship as a webserver and tell users to connect to via localhost. https://github.com/Asmod4n/hypha-mrb

t_mahmood 1 hour ago||

I am working on a minimalist journal app that is for really quick single line jots. Can be used for idea dumping, project management, quick calculations, unit conversion, task management etc. Have unlimited undo/redo support, assign tags etc. Will be adding scripting support using Python next. The data is stored in text file, in really human friendly manner, but also in way so that the *nix tool users can easy to navigate the file using text processing tools.

Native application, no web UI, built using Rust + iced.rs, minimal dependency. NO AI.

I am putting the best effort to make it performant. Target audience is the users who want's the simplicity of the notepad [non-sloppy one], but still with some bells and whistles to note without worrying about managing the metadata manually.

I think with scripting there will be infinite possibilities to play with linear notes, and I want to make that happen.

Continuous challenges while implementing features are:

  1. It should load instantly
  2. Keeping it extremely simple to use
  3. Keeping the interface minimal
  4. Still have ways to let the user find the features easily.

Will have a demo version ready soon

dielll 5 hours ago||

A small game called Colorhunt to play with my friends https://colorhunt.lol

You get assigned a random colour and have 24 hours to go out and take photos matching it. The game then generates a photo grid from everything you captured.

Modes:

- Solo → 9 photos by yourself - 1v1 → compete against a friend, combined grid at the end - Squad → everyone contributes to a 20-photo grid

No accounts, no app install, no personal data stored. Photos and generated grids auto-delete after 24 hours from Cloudflare R2 storage.

Made it for fun to find a way to do a shared a activity with my girlfriend and also to challenge my friends over the weekends.

Currently redesigning the frontend flow but I am kinda poor at design

Would genuinely appreciate feedback, ideas or anything

mstaoru 16 hours ago||

[NO-AI]

Being a weightlifter for 20+ years now, I'm working on a barbell speed and path tracking sensor based on newer IMU hardware technologies, which makes it both more precise and cheaper than camera- or actuator-based systems. Ultimately it helps you lift and train safer and better.

It's an intersection of industrial design, hardware, firmware, and software (and some sport science, of course). This intersection is not yet dominated by LLMs so it's a breath of fresh air.

In an early prototype stage as in "strap a Raspberry Pi to a bar", but it looks promising and I'm happy to move forward, also using connections from my previous 12+ years in China.

utopiah 4 hours ago||

How about strapping the phone to the bar and opening a Web page with https://developer.mozilla.org/en-US/docs/Web/API/Acceleromet... ?

Seems it would have a much higher reach.

mstaoru 4 hours ago||

Phone accelerometers don't have enough range and sampling frequency to even begin with. Raw, on some rare phones you can sample 800 Hz (enough-ish), but on most 100 Hz max, Web API is capped at 60 Hz, this is all way too low for any quaternion math. They also have much higher noise density which is the silent killer of all kinds of IMU navigation.

I also wouldn't trust a strap to drop a loaded bar from snatch :D https://youtu.be/nrgnH9fTfGo?si=6LLeu3y02iFrwfis&t=65

utopiah 2 hours ago||

Ah didn't think you'd need that much, thanks for the clarification.

Might consider a BT GadgetBridge gadget then.

gozzoo 3 hours ago|||

I've had the same idea for year. When google released their Fitbit Air few days ago I the first thing I tought was - can it be used as a sensor for weightlifitng and do they have API for that.

era86 16 hours ago|||

Wannabe powerlifter here of about 20 years as well. This sounds like an awesome project! Is bar-path the main metric for safety and "better" lifting? A project I had in mind, once upon a time, was an automatic "Form Check Friday" for myself using a Pi + Webcam.

cleaning 11 hours ago|||

As someone who has been very deep down this rabbit hole and hacked together multiple path and velocity trackers over the years (specifically for olympic weightlifting), there is no extra information that tracking bar path will give you that simply looking at the video won't, and often just adds more clutter. You don't need to graph bar path to see that the bar is looping too far forward after hip contact in the snatch.

Velocity on the other hand is a great metric to track and is used as a proxy for RPE. Mike Tuchscherer was the first one to systematize it for powerlifting a while back, if you've been lifting for 20 years you're probably aware of the name.

mstaoru 14 hours ago||||

Thanks! I think for "canonical" lifts (squat, deadlift, row, to some extent military press) the vertical bar path is mathematically optimal, and for all kinds of lateral or sagittal movements you do more work with weak stabilizing muscles and load joints laterally too. Is it productive work that strengthens your core? Possibly, but it's hard to quantify. It it something that can lead to injury? Absolutely yes.

For more complicated lifts like bench press (J-shaped) or snatch (S-shaped), for example, I would rather set a "golden sample" path with a coach and compare to that.

It's unlikely to be the sole metric, especially given the inverse kinematics of different body types (long/short femur, etc), but together with bar speed, over time, it can provide a lot of good feedback.

cleaning 11 hours ago||

It is not "absolutely" something that can lead to injury. Injury itself is difficult to define, and often the reason one experiences pain sensation is multifactorial. Within lifting contexts, generally the factor which has the strongest evidence for injury prediction is how sharply an athlete increases intensity compared to what they have previously adapted to.

No offense, but this post does come across as you only having a surface level understanding of the field. Especially surrounding injury/pain perception, I would be more careful of what you assume is true, there's far more nuance.

notesinthefield 15 hours ago|||

Not OP but velocity is typically what these devices are used for. Its a great measure of between-set intensity.

grepfru_it 12 hours ago|||

Side note: My LG Watch Sport smartwatch was able to determine what weight training workout I was performing and somehow figured the weight with astonishing accuracy.

a-dub 14 hours ago|||

i'm curious about how effective path tracking can be in comparison with computer vision based inverse kinematics of the body itself. do all forms of bad form have detectable imu signatures?

i wonder if it would make sense to consider it as a data problem, capture a bunch of high fidelity inverse kinematics data for various forms of bad form/dangerous lifting along with the imu data and then work from there. there could be some interesting and unexpected features that are easier to detect than straying from straight line paths with some tolerance.

mstaoru 5 hours ago||

For me it's a bit of an inverse problem. I go to a public gym (hard to sustain motivation at home) and I absolutely don't want to film myself there.

sberens 11 hours ago|||

Have you seen https://fort.cx?

frankdenbow 16 hours ago|||

this sounds awesome. have any videos of it?

mstaoru 14 hours ago||

Thanks! Working on it, for now it's literally a taped breadboard. :D

wrer 11 hours ago||

[dead]

markmnl 6 hours ago||

Working on fmsg - open protocol for instant messaging, distributed by domain like email..

Its a message definition and protocol, addresses look like @user@domain, anyone can run a host, and threaded messages are linked by cryptographic parent hashes..

The idea is to take the best from email: open protocol, domain ownership, interoperability (unsolicited mail is a feature not a bug), and the best from closed instant messaging rebuilt: efficient binary messages, conversational threads, sender verification, message integrity etc. built-in. Originally envisaged for human-to-human messaging but partculalrly interesting time right now with human-to-agent and agent-to-agent messaging...

The OSS stack is up and running: Go host, Dockerised full setup, CLI, Web API, and a spec nearing v1.0. Did Show HN post week ago: https://markmnl.github.io/fmsg/show-hn.html

Seeking feedback, criticism, validation :) protocol bikeshedding, and especially interest from founding-engineer types who want to help build an open messaging ecosystem rather than another closed app..

tgsovlerkhgsel 4 hours ago|

What is the unique selling point that it has over Matrix?

Matrix addresses have a similar format, anyone can run a host, open protocol, domain ownership, interop... Threaded messages are supported AFAIK, the details of the crypto will be different but overall it feels like it is close enough that a new protocol will have a hard time having enough advantages to overcome the huge network effect (Matrix being one of the few open messengers that actually have some following already).

markmnl 2 hours ago||

You're absolutely right to pick up on that, I did study the landscape and Matrix is closest.. biggest difference is fmsg is just messages - group like chats evolve naturally in the threads - but to get a message someone has to send you one. Group messaging platforms like Matrix, Rocket.Chat etc have concept of rooms/forums/channels i.e. groups, then have HTTP APIs to manage membership and sync messages.. fmsg just messages someone has to send you

Also fmsg being its own protocol can do novel things like to auto challenge during sending back to sender - can't do that with HTTP

shinrak 15 hours ago||

I finally managed to finish a project and publish my first game on Steam: https://store.steampowered.com/app/4195030/balls/

It's a short chain-reaction game in which you explode balls bouncing in the screen, and need to build up to target scores. You build bigger and bigger combos as the game progresses.

It was a blast to work on it, starting with a small toy and just adding features that "felt right" until I had a game that was fun to play. It was quite hard to find a balance though, so a lot of numbers are arbitrary - but I enjoy seeing people breaking the game in new ways and finding new builds.

These days I've been working on patching reported bugs and sharing the game with people. Now after the latest patch, I feel like I'm done, but I feel like going back at it and adding an idle mode. And maybe simplify the codebase so I can test and iterate better, and then add many more ball types...

I know that any good LLM could replicate this pretty quickly, but I made this myself and I'm still feeling proud of the accomplishment :)

markmnl 6 hours ago||

Shiping a real product, esp. a game, from scratch is an awesome achievement, congrats. (My first shipped product was Square Heroes also on Steam).

I feel like perfecting something can be trap, sure keep it alive, but maybe think about the next thing to work on too?

ditchfieldcaleb 10 hours ago||

Oh, this is cute and a great first game! I'm working on my first game as well (a top-down 2D tower defense game).

What engine or framework did you end up going with? I looked into Unity, tried Godot for a few weeks, but landed on just making a Typescript-powered canvas game with PixiJS for graphics rendering. Found it much easier doing it this way instead of having to learn a game engine.

bramadityaw 1 hour ago||

I am working on blase (https://github.com/bramadityaw/blase), a language server for Laravel's Blade templating language. This is a proof-of-concept project that I plan to make into a submodule inside a more general PHP language server.

wpietri 10 hours ago||

The last couple years I've been prototyping a not-for-profit pinball museum in Chicago. In the coming weeks we'll be opening 7 days a week in a 2900 sq ft space in the Loop: https://theflip.museum/

It's my first time starting a physical, retail business and it has been quite an education in the small details.

tehlike 6 hours ago|

Impressive. I hope you find success!

jfim 1 hour ago|

Building a personal internet archive and knowledge management stack. It's in a rough state but I'm using it every day.

Cham (https://github.com/jfim/cham) is an archive for internet content, you give it an URL and it'll archive it for you, extract the text with readability if it's an article, or extract the audio track then transcribe it. Content is automatically summarized and tagged, and you can start a conversation with a LLM about the article. It supports feeds too so you can subscribe to blogs and keep the articles in case the blog goes away. I still need to add search, improve the CLI, add all the missing features, and do a lot of improvements all over the place.

To improve reliability, I made passe-partout which is basically a Chrome browser with a rest API (https://github.com/jfim/passe-partout) and veilleur (https://github.com/jfim/veilleur) which turns any blog listing into a RSS feed. So this way I can take blogs that are rendered using JavaScript, don't have a RSS feed and load the articles directly into Cham.

Also built a modular MCP server with OAuth2 dynamic registration so that I can have my own MCP server that works with the web, desktop, and cli versions of Claude/Claude code. Currently have modules for editing files so that I can edit/search my Obsidian vault from Claude, fetching pages through passe-partout (since some pages block LLMs from reading them), and proxying MCP servers so that servers that only support bearer token auth can still work with web Claude.

Also, a gnome terminal emulator UI with some unique features like split browser/terminal tabs. https://github.com/jfim/jfterm

Mostly an excuse to see how far I can push LLM code generation to write tons of software that I've always wanted but never had the bandwidth to tackle, and learning to deal with the sometimes questionable code quality that comes from it.

More comments...