Ask HN: What Are You Working On? (March 2026)

Mostly Jolteon (https://github.com/lautarodragan/jolteon), a TUI music player written in Rust (for almost 2 years now!)

Also used the new Navigation API (and some Shadow DOM) to build a cheap, custom client-side rendering (sort of) into my site (https://taro.codes), and some other minor refactors and cleanup (finally migrated away from Sass to just native CSS, improved encapsulation of some things with Shadow roots, etc).

I've been wanting to write a simple AI agent with JS and Ollama just for fun and learning, but haven't started, yet...

DesignFlo:

https://designflo.ai

Build enterprise grade applications (in Elixir) with AI the right way.

Secure. Scalable. Reliable.

Built based on a senior engineer's experience. Uses 10 years of battle-tested patterns, not just LLMs:

1. Uses algorithms over AI whenever possible.

2. No external library dependencies whenever possible.

3. Old school over shiny new toys. Use the right solution for the problem (Eg. SQL vs NoSQL).

I'm building open source homebrewing (as in beer) software at https://www.brewdio.beer. It's something I've poked at periodically for a few years but now I'm using AI to see how far I can take it.

It has a few core libraries built in rust with a web app and a terminal UI. Android app is in the works. The persistence layer is intended to be offline first using a CRDT with an optional sync server. I'm also trying to integrate "bring your own AI" assistants to help tweak recipes or make suggestions.

It's been a fun way to sharpen my claude skills but also to see how feasible it is to maintain multiple frontend applications with a large amount of shared code. Still a lot to do, particularly the core calculations are not yet on par with existing offerings.

awesome. we have an official one in nyc. https://tree-map.nycgovparks.org/tree-map

learning how to fine tune image models, for an attempt at getting diffusion to output LWIR fire mapping data from RGB picture images

so far, ive spent a lot of manual time labeling and matching RGB and LWIR images, and trying to figure out first ways to get better pose matches when the flights arent the same.

that, and many different attempts at getting torch to work using my laptop's GPU and NPU. i think im close, without having to build torch from source woo.

Ive been having an eye towards getting better llm generation quality for python too, but havent put a focus on it yet. im fed up with it making one off script after one off script and instead of just making a react app, making some raw html and making a new html file with the new and old bugs every time i want to do something interactive. its maddening.

my last month of gettin claude code ro play pokemon webt well and ive about learned skills pretty well now, but it keeps wanting to do like a challenge run of sticking with a single pokemon.