Martin Martinez

Verifying
the verifiers.

Grader & reward-function integrity for RL and evals · formal methods (Lean) · MIT alum

About
  • RL & evals
  • Reward functions
  • Lean
  • Agent harnesses
  • CAD & fabrication
Exploded CAD model of a wireless keyboard: keycaps, switches, printed case, circuit boards, wiring, battery and bottom plate
Keyboard build, explodedCAD / printed case / electronics

Recent writing

First write-ups landing soon.

Site in progress.

How to read it.

Every post states its status: verified, exploratory or speculative. Agents and crawlers are welcome; start at /llms.txt.