Portfolio · Résumé · LinkedIn · Email

I tend to start with a product problem and end up several layers down the stack.

At Peptide AI, I was one of two engineers, so the work rarely stayed in one box. I rebuilt parts of the iOS and Android experience, shipped full-stack features, worked on the data and API layer, added product analytics, and automated weekly customer emails. Along the way, I built a privacy-scoped semantic cache that brought cached responses from 9 seconds to under 3. The app reached 14,000 downloads and 9 million store impressions in two months.

That experience shaped the kind of software I like building now: products where the model is only one part of the system. I care about what happens around it—how data moves, how behavior is evaluated, how failures become visible, and whether the whole thing is actually useful to someone.

I've been contributing to Inspect Robots, an open-source framework for evaluating computer-use agents. My first contribution turned operator-verdict normalization into a tested public API (PR #415). I also sent the maintainer a proposal for a backward-compatible metrics layer so evaluations can be observed and compared across runs. That part is something I want to build, not a merged feature yet.

I'm also interested in agents that have to do real work instead of only returning an answer. That led me to build an eSource study builder that configures clinical-study forms across unfamiliar web platforms, and a provenance-aware extraction pipeline for schedules buried in difficult protocol PDFs.

I still enjoy building outside that lane. Sometimes that means a full-stack simulation tool like SynthDrive; other times it means implementing and testing the lower-level pieces myself, as I did with the convolution backward pass in my AI image detector.

Most of my work ends up somewhere between TypeScript and Python, React and mobile, APIs and Postgres, with evaluation code nearby. I care more about understanding the system than collecting frameworks.

I'm studying Computer Science at the University of Maryland and graduate in May 2027. I'm looking for a team where I can keep working this way: close to the product, willing to cross layers, and serious about turning ambitious ideas into software people can use.