about
One workshop. Many notebooks.
KeiLab is a single-founder research lab building mathematical foundations under modern AI — from physics framework to hardware silicon.
The thesis
Modern AI runs on linear-algebra primitives whose interpretation is largely empirical. KeiLab develops an alternative — a math-first framework where the same operator structure that drives a transformer attention head also drives a quantum measurement, a biological gene-expression dynamic, and a control loop in a prosthetic limb.
The framework yields three layers of output: theory → patents → applied hardware. All three layers run on one workstation, are maintained by one founder, and emit reproducible artifacts on a local version-control server.
How it builds
Each derivation is a markdown file. Each formula gets a falsifier. Each theorem closes against a registered observable. Failures are documented as honestly as successes — the public catalogue of refuted hypotheses is part of the lab's record.
The lab's self-hosting compiler, agent runtime, and GPU training stack are all open-source-by-default for the parts that do not carry patent-sensitive methods.