The record behind the practice.

I architect enterprise AI for a living. At Thomson Reuters I direct the global systems architecture for the enterprise Generative AI and multi-agent orchestration platform serving legal, tax, and financial products — the environment where a wrong answer is not a bad demo, it is a liability.

That constraint produced a conviction: a single model returning a single confident answer is the wrong shape for executive decisions. Verification is not consensus. So I build panels that disagree on the record, with humans holding the pen at every high-stakes gate.

As CEO of Compound Talks, I run that same architecture in public — two synthetic executives reading the same headline from opposing desks, twice a week. It is not a content project. It is the advisory pattern, operating where anyone can audit it.

The work extends to policy. Designated by the National Center for State Courts and the Thomson Reuters Institute, I consult with Supreme Court justices and federal court administrators on secure AI frameworks for judicial operations — hallucination mitigation, model parameters, private inference. When the institution cannot absorb a fabricated citation, governance stops being theoretical.