I am a software engineer at Sixth Street, where I build data pipelines and reporting tools for firmwide metrics, working directly with the C-suite and finance teams. Before that, I interned at Sixth Street on API and full-stack work, and at Trace3 on data engineering and cloud migrations. I studied computer science and statistics at UVA. I also founded Pitch, a social music app on the App Store, and co-founded Picket. These days, I spend most of my free time building AI agents and evals, digging into how they work, and writing about what I learn.
An agent eval harness. A LangGraph agent audits financial models for hardcoded values, off-by-one ranges, and unit mismatches, using tools to read formulas and run Python. A bug injector and scorer measure precision, recall, and false positive rate, with LangSmith tracing, and I compare models on accuracy and cost per audit.
A personal finance dashboard that aggregates all of my bank and card accounts through the Plaid API into a single view of balances and transactions. An agent runs on every new transaction, evaluates it against my budgets, and sends me a reminder when I'm close to or over a limit.
A social music discovery app where friends share weekly song picks. Expo React Native and TypeScript on the front end, Django and PostgreSQL behind it, with real-time features over WebSockets and Spotify, Apple Music, and Deezer integrations. LLM-generated personas power discovery.
A mobile social app built around making decisions together. I built the Django REST backend and push notifications through Expo, and designed the interface in Figma.
Writeups of my own work: the problem, how I built it, and what the evals showed.
one paragraph abstract
what goes wrong in financial models, and why it matters
agent design, tools, and how the bug injector works
dataset, metrics, models compared, and number of runs
precision, recall, false positive rate, and cost per audit
what it misses, and what I would try next
My thoughts on all things tech and life.
post text