Work
Production AI inside a global compliance product, an enterprise platform migration that held customer trust, and the architectural patterns I prove out before bringing them into regulated environments.
A graph database with explicit, named musical evidence on every connection: the architecture for steerable, explainable recommendations. Validated on music; portable to compliance, where regulations, substances, and enforcement actions have the same relational shape.
An architectural pattern for running external AI models against sensitive data without the data ever leaving the trust boundary. Tested on personal finance; the same shape applies to medical, legal, HR, and regulatory workflows.
An automated pipeline that normalizes environmental enforcement actions from 23 regulatory sources across 16 jurisdictions into a single queryable compliance dataset, with AI-imposed structure and a clean REST API.
An AI customer-intelligence system that queries 74,000+ recorded calls and 2.5 million transcript sentences. Changed how the product team prioritizes - from gut feel to a queryable source of truth.
A production AI system that replaces a manual auditor process. Takes a facility profile, runs 500+ screening questions in batch, returns structured applicability decisions with reasoning auditors can act on.
Half the enterprise customer base, off a legacy platform, onto a modern architecture, in a single year. The project that proved we could run a complex high-stakes delivery program without dropping things.