AI Architecture · 2026 · Active
An AI food discovery platform
Search that answers what to eat, not just where. Built end to end.
My part
Retrieval and ranking, backend and data platform, web, iOS, infrastructure
This one I cannot name yet, which is annoying, because it is the hardest thing I work on.
The problem is simple to say and unpleasant to solve. People do not want a list of restaurants. They want to know what to eat. Answering that means treating a menu as structured, searchable data, and menus are not structured data. They arrive as scanned PDFs, as photographs, as HTML written in 2011, in several languages, with ingredients spelled four different ways on the same page.
Everything hard about it follows from that.
If someone filters for an allergen, the answer has to be right. Not mostly right. A search that is usually correct about peanuts is worse than no search, because people stop checking. So the interesting question is never "can a model extract ingredients", it is "what do you do with the extraction you do not trust", and how you keep that judgment consistent across a catalogue that grows every day.
I build it end to end: the retrieval and ranking, the backend and data platform, the web app, the iOS app, and the servers it all runs on. I work with a team.
That is as much as I can say about it for now.