Most AI assumes a data center. These assume a pocket, a breadboard or a browser.
Plain language in, tool calls out. A tiny model that runs fully offline on an ESP32, built for makers wiring their own devices.
“turn the fan on if it gets hot”
{ "tool": "relay.on", "when": "temp > 30" }
An LLM inference engine written from scratch in C++. No llama.cpp underneath. Aimed at the budget Android phones most people actually own.
A tool calling model under 100 MB. Grammar constrained decoding means every call it emits actually parses.
Selective about projects. Ambitious about the ones I take: models, engines, and the tools around them.
Start a project →benedictpatrickjohn@gmail.com