Broccoli AI
Building agentic software for home-service teams: technician chat, dispatcher workflows, search, data pipelines, and operational tools.
AI product engineer in San Francisco
I turn messy workflows into fast, reliable software.
At Broccoli AI, I build agent-driven tools for home-service operators: chat surfaces, search, workflow automation, data pipelines, and the interfaces that make them usable.
Before that I co-founded Serra, a YC S23 recruiting company, and did bioinformatics work at Quon Lab. I care about systems that are practical, fast, and easy to trust.
Work
A few places I've built product, infrastructure, and AI systems.
Now
Serra
Co-founded a YC S23 recruiting platform focused on search, matching, and workflow automation for hiring teams.
Quon Lab
Worked on bioinformatics research at UC Davis, where I learned to care about data quality, reproducibility, and complex information systems.
Notes
Short ideas from things I've actually been building.
Working noteAgents need boring infrastructureStreaming, resume, persistence, evals, and the unglamorous parts that make AI tools feel reliable.Working noteSearch is product designEvery workflow seems to need a different kind of search, and generic search often misses the point.Working noteData plumbing is product workA few notes from replacing brittle integrations with simpler syncs, raw tables, and useful internal analytics.
Misc
A few recurring interests.
Agents for field-service work, better internal tools, and interfaces that make operations less chaotic.
TypeScript, React, Python, Postgres, ClickHouse, search, evals, and enough infra to keep product work honest.
SF live music, tiny web tools, black cats, and overly specific data-cleaning projects.
What changes when software can keep context, take action, and still stay understandable?