Curiosity,
with a paper trail.

This is my open notebook for local AI: language models, quantization, speech, images, coding agents, and policies that learn from pixels. Some entries end in a useful deployment. Others end in a rollback, an incident, or a better question.

I want the part between “the model fits” and “the thing works” to be visible. That means actual requests, recorded outputs, resource measurements, and acceptance gates that can say no.

The editorial starting point is inspired by The Kaitchup — AI on a Budget: a practical question, a concrete experiment, and the result. The stories and measurements here come from my own experiment archive.