I lead data product teams by day. This site is what I do with the other hours: Court Vision, an attempt to teach a computer to chart tennis matches from broadcast video.
The Match Charting Project has volunteers hand-coding every shot of professional matches — one of the best open datasets in sports, bottlenecked by human hours. Computer vision models like Meta's SAM 3 can now find and track a tennis ball from a text prompt. This project is me finding out how far that actually goes: from raw broadcast footage to point-by-point match data, built in public, dead ends included.
Before data products I spent time in economics and sports analytics, and I never stopped thinking about measurement problems — what the data actually tells you versus what people want it to say. Charting tennis automatically is the purest version of that problem I've found: the ground truth is right there on the screen, and the entire challenge is turning what you can see into data you can trust.