Experience, projects, and education
Where I worked
I sat shoulder-to-shoulder with functional subject matter experts across Legal, Editorial, Engineering, Operations, and beyond to deeply understand their workflows, then designed, prototyped, and delivered the technical solutions that changed how those teams operate.
Directed program management for software, process, and AI automation initiatives across 4 teams in digital healthcare advertising. Acted as the liaison between engineering and business stakeholders, and owned delivery from requirements through production handoff.
Built and delivered analytical systems for government clients. Owned requirements through production, and the accountability that came with them.
Built a proof-of-concept AI meeting-transcription app. First production code I wrote.
What I've shipped
I went back to my 2018 Johns Hopkins master's project and found it had never actually computed a sound source's position, just an unreduced cloud of sample points. Audited the code, found four silent bugs, rebuilt the missing triangulation math from scratch, and measured exactly where the physics runs out in a reverberant room.
Campaign finance and voting records for every LA city official, side by side. Started as a one-day build at Anthropic's LA Claude Impact Lab covering three officials, then extended to the mayor and all 15 council districts, with a donor contribution map and a sourced "Ask AI" box on every official's page.
AI coding tools generate deployment configs that our internal platform silently discards. I got one application deployed end to end, worked backwards into a Claude Code skill pack carrying the platform's real constraints, and submitted it. The platform team adopted and extended it. I then found and fixed the bugs their extension introduced.
BuildDiff is a deterministic comparison platform that checks regulated web content across six layers — text, styling, markup, images, layout, and tracking links — and presents the differences through Visual, Source Code, and Overlay views. It replaced the legacy QA tooling and the manual review passes around it for copyeditors, QA validators, and front-end developers at Medscape. The comparison engine is fully deterministic, with an AI layer sitting on top read-only for triage and summarization, never as the source of truth.
Rolled out an enterprise AI platform to 100+ active users across seven business units, then built the seven workflows that made it stick. Highlights include an autonomous M&A due-diligence agent that replaced a six-week manual process with a 3-hour run.
My first agent system. Nine agents automating campaign operations at WebMD / Medscape: file reprocessing, delivery auditing, engagement tracking, compliance validation. Where I learned that the hard part of running agents in production is rate limits and partial state, not prompting.