I build systems that hold up — databases, platforms, and now training/evaluation environments for AI agents. The question underneath all of it is the same one: how do you know a system, or an agent, actually did what it claims?

Currently building Crucible — turn any real software (a database, a CLI, a codebase, an API) into a trainable, gradable, replayable RL environment. The reward is your real software; every episode replays byte-for-byte, so every reward is auditable. MIT.

→ A 0.5B model went 5%→100% on a real SQL task with a Crucible environment as the only reward — then generalized across shell, code, and database agents. The proof.

Also building Loam — a moddable engine for growing your own AI-inhabited world.

Day job: database platform engineering at scale — hundreds of Postgres/MySQL instances on AWS, increasingly AI-assisted: production automation that reviews real database changes before an engineer sees them.

How I build: one source of truth, reuse over rebuild. Tests ship with the change. Docs move in the same commit as the code. One production path — no MVP-vs-real branches.