I compile research into production systems.

Most days I live at the intersection of messy real-world data and models that have to work anyway — turning fuzzy inputs into something reliable, teaching machines to recognize when two things are really the same thing, and building pipelines that hand off clean output to whatever comes next.

The loop: experiment → benchmark → ship → repeat. I care about the gap between “works in a notebook” and “works at 2am when nobody’s watching.”

Off the clock: CTF puzzles, RC builds, gym PRs, and rabbit holes into how things work under the hood.