I build production LLM systems and find where they bleed money. Started in warehouse ops, taught myself to code to kill the manual work my team was drowning in. Now I ship agent systems, evals, and data pipelines, built to run cheap. Cost discipline isn't a feature I bolt on, it's how I build.

Live at claygeo.dev

Flagship work

competitive-intel-platform — Production data engineering at scale. A 14-state, 65+ store competitive pricing pipeline: reverse-engineered 3 proprietary retail APIs under Cloudflare with Playwright, a BullMQ + Redis worker fleet, Postgres normalization, OCR via Google Vision. Replaced a 2-week manual cycle with same-day data. Used daily by my pricing team for over a year.

Claude/Codex Plays — Autonomous agent that plays Pokémon end-to-end from RAM state, adversarially gated so it can't fake its own progress. Agent orchestration, eval-harness design, and honest failure logging. Clips and writeup →

How I work: I ship end-to-end, to production, measured. When one project's exploit rate dropped from 67.7% to 13% on a random sample, I published both numbers and treated the gap as a design problem. I write things down honestly.

Stack: TypeScript · Python · Rust · SQL · Anthropic / OpenAI / OpenRouter / Ollama · eval harnesses · agent orchestration · Postgres · Supabase · Redis · Playwright · Next.js · Netlify / Vercel / Hetzner

More at claygeo.dev · LinkedIn · X