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

I find it fascinating to give the LLMs huge stacks of reflective context. It's incredible how good they are at feeling huge amounts of csv like data. I imagine they would be good at trimming their context down.

I did some experiments by exposing the raw latent states, using hooks, of a small 1B Gemma model to a large model as it processed data. I'm curious if it is possible for the large model to nudge the smaller model latents to get the outputs it wants. I desperately want to get thinking out of tokens and into latent space. Something I've been chasing for a bit.

A native WebGPU JS engine (no browser needed) https://github.com/mystralengine/mystralnative/

Already have my own JS engine & the basics of three.js and pixi.js 8 working, roadmap to v1.0.0 posted in github issues. Aiming to show it to folks at GDC in March.

Borrow This And Improve It - an app for tracking repairs to a thing (right now it's bicycles only, planning to extend to other things like leaf blowers and electronics) and giving away half broken things or repaired things to others but with the advantage of also giving away the repair history for something. So for example, I found a bicycle in the trash, fixed it up with a few new parts (less than $6) and soon I'll try to give the now repaired bicycle away to a new home, plus a QR code that links to its repair history. The idea being that knowing how something was fixed once will make it more likely that it would be fixed again.

How Home Alone My House - A fun app I'm making with my children using computer vision. The idea is I can scan the room with my camera before unwittingly walking into their traps and becoming a hapless adult who didn't pay close enough attention to tripping hazards and choke lines.

I’ve been experimenting with a live win probability predictor for the 10-player arcade game Killer Queen. The goal is to predict the winner in a causal, event-by-event fashion.

Right now I’m struggling to beat a baseline LightGBM model trained on hand-engineered expert features. My attempts at using a win probability head on top of nanoGPT, treating events as tokens, have been significantly worse. I am seeing about 65% accuracy compared to the LightGBM’s 70%. That 5% gap is huge given how stochastic the early game is, and the Transformer is easily 4 OOM more expensive to train.

To bridge the gap, I’m moving to a hybrid approach. I’m feeding those expert features back in as additional tokens or auxiliary loss heads, and I am using the LightGBM model as a teacher for knowledge distillation to provide smoother gradients.

The main priority here is personalized post-game feedback. By tracking sharp swings in win probability, or $\Delta WP$, you can automatically generate high or low-light reels right after a match. It helps players see the exact moment a play was either effective or catastrophic.

There is also a clear application for automated content creation. You can use $\Delta WP$ as a heuristic to identify the actual turning points of a match for YouTube summaries without needing to manually scrub through hours of Twitch footage.

You can do it!