How this lab happened
One loop, closed five times.
Every project here is the same shape: a system that observes its own output and corrects itself. The lab itself was built that way — each stage produced the tools the next stage needed.
The trade
Industrial electrician. Control loops, fault-finding, and systems that fail safely — the instincts everything else is built on.
MirrorBot
A conversational-safety layer for a live Discord community — built on intuition and iteration, deployed for over a year across tens of thousands of interactions. Its real product was the questions it raised.
T³
Those questions, formalized: a Clifford-algebra transformer architecture and a multi-year research program on self-regulating computation — multi-seed, preregistered, under peer review.
The mesh
A six-node compute cluster that routes its own context, monitors its own training runs, and heals its own services. ~78K lines of infrastructure, 9 systems in production.
The fab node
The loop goes physical: agent-designed parts, printed and camera-inspected autonomously. Its first shroud dropped a training GPU from 81 °C to 61 °C.
MirrorBot deserves an honest word. It was built in my pre-rigor phase — instinct-driven, informally evaluated, and deployed live because that's how I learned. It worked well enough to run for a year, and it tracked regulation signals that turned out to matter. But its lasting value wasn't the deployment; it was discovering, firsthand, what informal evidence can't prove. The T³ program exists because I wanted answers that would survive someone else checking. That transition — from a system I believed in to claims I could defend — is the actual origin of this lab, and the reason rigor is in the name.