Tom Sullivan

I build systems that must be right.

Fifteen years turning enterprise data and AI from strategy into shipped, governed systems, across financial services, insurance, and life sciences.

Built

• model-eol: A machine-readable deprecation feed format for AI models, and the zero-dependency Dependabot-for-models that runs on it. npx model-eol . --days 90
• adaptive-academy: A local-first learning app. The tutor won't give you the answer until you've tried.

Contributed

• NVIDIA-NeMo/Switchyard: two merged fixes. Resolve routing tier for fallback and affinity-reuse decisions. Re-emit captured provider extensions when encoding to the Responses format.
• zenml-io/kitaru: one merged change. Support Node.js 24 in TypeScript packages.
• openai/codex-plugin-cc: one open pull request. Unsubscribe task threads after client disconnect.

Led

• 75+ engineers led across enterprise teams
• $317MM+ realized and projected value: $52MM realized cost avoidance + $265MM projected 3-year profit from ML
• 1.1M documents a year processed at 80% lower cost

Writing

• Forward-Deployed Engineering Is New. The Work Underneath It Is Not.
• The Field Guide
• The Rest of the System, the newsletter

tomsullivan.dev · LinkedIn