- 🤖 Senior AI Engineer at Fingerpaint's Wetpaint Innovation Lab, building LLM-powered tools for regulated pharma
- 🧱 Full stack, in the literal sense — blank repo to production. React and TanStack on the front, Hono and FastAPI services behind it, Postgres and Snowflake under that, SSO / passkeys / multi-tenant RBAC over the top, Docker and Terraform holding it up, and the agents and evals running through the middle.
- 🧪 Most of my time goes to evals — LLM-as-judge alignment, feedback pipelines, and dragging evaluation out of the codebase and into the hands of the people who actually own the domain
- 🚦 Firm believer in wrapping non-deterministic agents in very deterministic gates
- 🐍 TypeScript and Python, mostly
- 🤙 Always looking to connect with like-minded devs!
- An agent that ships code — watches Linear tickets, hits an API I built, spins up an E2B sandbox, clones the repo, does the work, opens a PR. Every run has to clear lint, tests, and CI, and when it fails a gate it writes notes to itself that get loaded back in next time.
- An autonomous job-search agent — the whole vertical slice: ingestion straight from ATS APIs instead of job boards, a scoring layer for ghost-job likelihood and fit, a SQLite brain that lives in git, a React front end, and scheduled cloud runs that commit their own work.
- Evals for people who don't write code — the problem I keep circling back to, and as much a UI problem as an eval problem. Subject matter experts should be able to drive evaluation from an interface instead of waiting on an engineer to translate their judgment into a test.
Going deeper on fine-tuning, especially smaller models. I want the whole loop in one pair of hands: build the app, run the agents, evaluate them, then use what the evals tell me to tune the model — and back around again.
Front end
Back end & data
AI & agents
Evals & observability
Infra & platform
