I completed most of my last semester at the University of Michigan implementing Paxos and a sharded key-value store, writing CUDA kernels from matrix multiplication through the forward pass of a CNN, and replicating the vLLM KV cache optimization as an ML Systems researcher. None of it was needed to graduate. I took it because I wanted to go deeper.
Replaced a website that was an overengineered resume with something more genuine.
I entered an education hackathon with a team, twelve hours to build, and walked out with a fully featured platform for teachers to monitor student AI usage. Polished UI, real functionality, no smoke and mirrors. We went all in on the product and the polish, and not enough on the pitch. A CS team at a business hackathon, judged on a 90 second silent video and a slide deck. The platform worked. We just didn't prove that it mattered.
I built a test suite in Go and Python to find an edge in prediction markets, checking weather and niche sports markets for median price deltas. The edge was there until the bid-ask spread on thin liquidity swallowed it. Basic analysis, not exhaustive. Shelved for now, worth revisiting in a few months.
The job search was a numbers game. Final rounds across startups and national labs, moving in another direction from most of them. Then Google. Starting in San Bruno on June 1st. Walking at Michigan before that. Still figuring out what this site is actually for.