I like taking an idea far enough to find out whether it works. Sometimes that means building a model or simulation; sometimes it means benchmarking GPUs, reverse engineering a robot, or figuring out why a promising ML model falls apart outside a notebook. This site is where I write about those investigations.
For the past four years at Red Dog Media, my work has centered on production machine learning: causal inference, pricing experiments, and the platform work that gets a model from notebook to real decisions. That sits on top of a background in backend and data systems, and a degree in electrical engineering underneath all of it. I'm especially interested in the gap between a model that looks good in evaluation and a system that makes useful decisions in practice.
Outside work, I follow questions across math, software, and hardware. I've explored the constraints of a hexagonal puzzle, simulated competing advertisers, and tested local language models on a three-GPU machine. I enjoy projects where building something changes my understanding of the problem, even when the original idea turns out to be wrong.
I'm interested in work involving learning, optimization, and computing systems, especially where I can investigate an open question and help turn the answer into something that works. If you're working on something along those lines, I'd like to hear about it.