Product manager building software for federal student aid, where a wrong number becomes a compliance problem. 14 years in education finance technology. U.S. Army veteran.
At work I built my company's first AI assistant for financial aid regulation in Microsoft Copilot Studio and taught 100+ colleagues to build their own Copilot agents. Outside work I build open, tested tools for the same domain.
- Project SOR: an open calculator for the OBBBA Schedule of Reductions, with a staff tool, a student estimate, a free public API and an MCP server so AI agents query the real engine instead of guessing. Used by financial aid offices at 80+ institutions. 160 automated tests. Live
- financial-aid-documents: public federal source documents for an AI financial aid policy assistant.
A held-out evaluation for my AI reconciliation lab: new cases, an expected safe next step for each, blind human grading, and a comparison with the non-AI view, reported with cost and response time. Current results: evaluation results.
- Cite or don't claim. Every number traces to a labeled source.
- One engine, many views. Staff, students, APIs and agents use the same tested logic.
- Make it checkable. Public scenarios, contracts and a way to challenge any result.
- Ship the boring parts. Versioning, disclaimers and regression tests are the product.