# Gemma-based real-time inference project hunch is open-sourced by mmastrac

> Original title: Project by @mmastrac! He also open-sourced the approach as ‘hunch’ if you want to try it yourself:

- Source: Google Gemma
- Published: 2026-11-04T17:44:29.000Z
- Bad Influence: https://badinfluence.news/items/fea457rw21ot7jy18i72qlk0c
- Original: https://x.com/googlegemma/status/2118036677017153870

## Summary

mmastrac has open-sourced hunch, a project that uses EmbeddingGemma2 as a real-time /v1/systemone endpoint. The endpoint runs in about 20ms on text and 100ms on images, is trainable in real time, and can distill an upstream teacher model. The author says the approach is multi-modal by default and invites others to try it on GitHub.

## Post

Project by @mmastrac! He also open-sourced the approach as ‘hunch’ if you want to try it yourself:

Demo post: https://x.com/mmastrac/status/2115098609356284223
GitHub:
https://github.com/undrgrnd-labs/hunch

## Quoted post @mmastrac

> It's been a while since I showed any Jev stuff but I've been cooking on a way to use @GoogleDeepMind's EmbeddingGemma2 as an insanely fast real-time /v1/systemone endpoint (20ms on text, 100ms on images).
> 
> It's trainable in real-time, can distill an upstream teacher model, and is multi-modal by default. Hard to believe a @googlegemma model of this size can do so well with the right harness!

https://x.com/mmastrac/status/2115098609356284223