jax-gemini: build JAX/Flax models from plain English via Gemini

weskambale · x · 2026-11-20

An open-source Python library, jax-gemini (MIT, on PyPI), lets ML practitioners and domain experts build, train, evaluate, and snapshot JAX/Flax neural networks with conversational plain-English prompts instead of JAX boilerplate. Unlike descriptive LLM wrappers, it returns real, executable Python/JAX flax.nnx.Module objects — Gemini generates the code, which the library executes securely in a sandboxed environment.

Original post →

More from coding & agent

coding & agent channel →