I build evidence-driven AI products, readable agent systems, and decision tools.
- Agent System Mini — a small, inspectable agent loop with tools, SQLite memory, and traces.
- Point2Prompt — a browser pointing layer that turns selected UI elements into structured change briefs for coding agents.
- SplitTaste — mixed-household streaming recommendations with user-guided taste lanes and reproducible MovieLens evaluation.
- Where to Sit — a Seattle IMAX seat-view simulator for making a better seat choice.
- Microsoft Agent Framework — surfaced authorization URLs for Work IQ A2A-preview tool sources, with regression coverage.
- AO Bench — added a runnable comparison walkthrough for evaluating two agent adapters across capability and governance dimensions.
- AO Bench — merged a reliability fix for missing or empty benchmark-run reports, with regression coverage and a fully passing CI matrix.
Agent reliability, useful evaluation, human-in-the-loop product design, and tools that make complex systems easier to inspect and improve.