slvDev/esp32-ai ? reverse-engineered prompt

Build me a small open source project that shows a language model running directly on an ESP32 S3, with no server involved. I want it to generate short, simple stories on a tiny attached screen, and I want the repo to include the firmware, the training and quantization code, and a clear setup guide for wiring, flashing, and running it yourself.

Make it feel like a real research demo, not a toy, with code that can reproduce the model export and the on device inference flow. Include enough notes so someone can understand how the model fits in limited memory and why it works on such a cheap microcontroller. If you need current docs or example code for the chip or toolchain, look them up online.

Also add a concise results writeup so it’s obvious what was tested, what speed it gets, and what the limits are, especially that it can write stories but not do general chat or coding.