I make AI do what it’s supposed to, for the people who depend on it.

Led the behavioral testing behind California’s new AI assistant, AskCA

I designed the instrument: simulated residents, ordinary and adversarial, questioned AskCA, and AI judges scored every answer against its own instructions. Our method checks that AskCA answers from legitimate government sources and gives residents accurate answers. It’s part of the trust and safety framework ODI built with USDR and Stanford RegLab.

Quoted in Governor Newsom’s AskCA launch, September 2026: “Trust in a tool like this comes from testing it that rigorously, not from hoping it works.” Read the release → Also: ODI, “Lessons from the fire,” Sep 2026 →

Architecting a much loved retailer’s high-volume voice commerce agent

A voice commerce agent, and one of the most complex I’ve worked on. Its callers are loyal and particular, so the conversation has to flex the way a good salesperson does and still feel natural. I scope it, design and prototype the conversations, and build the structured tests that show it behaves before callers meet it.

Built a quoting engine to speed up a specialty insurance agency

Cole is a small agency writing specialized insurance. I built the custom software that does the routine work: a submission comes in, its coverage is checked, it is priced from the agency’s own rates, and the application is filled and queued for a person to approve. AI working for an everyday business.

Launching with the agency soon. I also trained their team to use AI day to day.

Building and shipping Otto, an AI agent for car shoppers

Every question clients pay me to answer — what the assistant says, when it stays quiet, what it will and won’t sell — I answer on my own product first.

Try Otto at otto2auto.com →