I build learning products, instrument them, and measure whether they work.
A data scientist and educator. Seven hundred-plus tutoring sessions taught me where people get stuck; the rest of my work is proving what moves them.
I studied mathematics and statistics at Reed, then spent 700+ tutoring sessions sitting with people while they worked. Most of them were not bad at math. They had one gap from years earlier that nobody ever caught, and once you find it and say it out loud, the rest moves fast.
The habit of checking started with my thesis. The model looked good on the surface and I felt fine about it for about a week, until I pulled the results apart and found it was failing the group I most wanted it to get right. Everything interesting was one level below the number I had been looking at.
At Yuno Learning I had to prove an intervention worked instead of assuming it had, which is harder and a lot more interesting. I came out of it trusting measurement over instinct, including my own.
Now I build learning products end to end, instrumentation included, so I can tell what actually moved someone and what only looked like it did.
Lately I've been evaluating language models, which turns out to be the same question in new clothes. Does the thing work, or does it only look like it works when nobody measures carefully.
Away from the screen I powerlift, play tennis, and cut linocuts, and I read more about automotive design than I can justify. They all have the same appeal. Form, patience, and the slow work of getting something right.