Healthcare by day. AI tools by night.
I've spent my career building products inside hospitals and health systems, most recently leading an Epic implementation at Memorial Sloan Kettering. In my own time, I build AI tools for friends and family, and I test them the way I'd test anything going in front of a clinician.
I'm always happy to talk about AI and healthcare products. Email me.
An AI agent that answers questions about NYC neighborhoods using live city data. Nothing is scripted: for each question, the agent decides which of five NYC Open Data sources to check, in what order, and says so when the data can't answer it.
What I learned: writing down what a good answer looks like before testing. My first eval run caught the agent calling an area "quiet" when the data was simply missing.
Claude tool use, NYC Open Data, Netlify Functions
A trip planner for following a Serie A club. It pulls the club's real fixtures and builds a day-by-day itinerary around the matches.
What I learned: working within real limits. Long trips ran past the host's time limit, so I switched to a faster model and checked that the quality held up.
Claude API, football-data.org, Netlify Functions
A place to log the fragrances you own and love, and get recommendations based on your taste. Each person's entries stay private to them.
What I learned: AI app builders are fast, but they need a close eye. I caught it quietly undoing a security fix, and now I check its changes myself.
Supabase, Claude API, Lovable
Before I test something, I write down what a good result is. It keeps me honest about whether a change actually helped.
I'd rather a tool say "I don't know" than guess. Missing data is not a finding, and an answer should say how fresh its sources are.
Most of what I build starts with a person I know and a question they keep asking. Seeing them use it is the real test.