Dev Fits Local Barista AI onto an $8 ESP32 Microcontroller

minchoi · x · 2026-08-04

A developer has successfully implemented fully local AI inference on an $8 ESP32 microcontroller. By typing a coffee-related question via USB, the model streams its answer directly to a small OLED screen without needing any cloud compute or GPU.

The project utilizes an asymmetric vocabulary design to drastically reduce the output head parameters, fitting the model into the highly constrained chip. This hardcore geek experiment demonstrates the interesting potential of edge AI in narrow, specialized use cases.

Related event: Developer Runs 28M Parameter AI on $8 ESP32 Microcontroller(2 posts)→

Original post →

More from Fun

Fun channel →