about
A research lab built for depth, not headcount.
KeiLab develops mathematical foundations under modern AI — from matrix-state theory to a bare-metal OS for real boards — with a tight team and a unified math-first language.
The thesis
Modern AI runs on linear-algebra primitives whose interpretation is largely empirical. KeiLab develops a math-first alternative — the Genesis matrix-state framework — aimed at routing: dispatch decisions and confidence gating for language models, tested against standard baselines instead of assumed to work.
The work has two layers: theory → the engineering that tests it — a compiler, an operating system, LLM inference and training, agent tooling, and hardware designs.
How it builds
Each derivation is a written document, and the aim is a falsifier for every claim — a test that would fail if the claim were wrong. Failures are recorded as carefully as successes; the catalogue of refuted hypotheses is kept in the lab's internal record and is available under NDA.
Engineering fleet
A local NVIDIA GB10 workstation for development and inference, plus rented NVIDIA datacenter GPUs for fine-tuning runs. Every cloud launch passes through a pre-flight memory and cost gate before the meter starts.
