Research
Build it. Check it. Find out what actually works.
Alongside building software, I work on questions about how we evaluate it: what counts as evidence, where automation helps, and what still needs a human check.
Berkeley Sky Computing Lab
Empirical software engineering · Coding agents
Ongoing work
What does an AI code review actually catch?
I study how coding agents review and revise software. The interesting question is not whether a reviewer produces a convincing explanation, but whether the alleged defect is real and the subsequent change fixes it. My focus is careful evaluation: connecting review traces to code evidence and keeping reported outcomes separate from independently checked results.
Sky Computing LabCAHLR · AskOski
Educational software · Applied machine learning
Ongoing work
Making research useful to students.
I contribute to AskOski, a Berkeley research and development effort around course guidance and exploration. My work connects backend engineering and analytics with the practical requirements of a student-facing product. I am interested in the distance between a useful recommendation in an experiment and a service people can rely on.
About AskOski