What I've Built

Started as a technical assessment, turned into ongoing work. The problem: turning messy healthcare PDFs into regulatory documents where the structured data has to be near-perfect. Tables, citations, cross-references — the parts where LLMs aren't reliable on their own. The AI handles the extraction and reasoning, but the final output is generated deterministically so it's reproducible and auditable. I also built the testing infrastructure so every section gets measured against expert ground truth.

A 20-person engineering team rebuilding core infrastructure needed to get organized about how they use AI during a major migration. I built the system around it: shared docs that stay current with the architecture, automated digests that pull key decisions out of team conversations, and a linked wiki that gives every AI session real context about the codebase. The work shifted from "help us use AI tools" to "build the thing that makes AI work for the whole team."

Built and shipped a production ML model that predicts optimal scheduling windows. Trained an ensemble on historical data, validated with back-testing, and deployed an inference endpoint running real-time predictions inside their app.

Designed and ran an all-day AI workshop for faculty at a large university hospital. Worked with their leadership to understand what they were struggling with, then built a curriculum around that. Hands-on exercises, follow-up videos.