The AI layer of a modern operation. I build agents that answer the phone, draft the reply, and chase the lead, then keep them monitored, measured, and honest in production.
What each line covers. Everything here has shipped, or runs in production today.
A Mac software company answers its public support numbers with a voice agent I operate. Product help, order status, troubleshooting, in multiple languages, with a warm hand-off to a human the moment the AI reaches its limit.
A consumer hardware brand drafts its support email through an agent I built. It reads the full thread, the order, and a versioned knowledge base before it writes, and it says "I'll check" instead of guessing. A human confirms every send; a hundred-ticket day stays a normal day.
AI Failure Analysis is where I lead. In live systems I have caught a cross-customer data leak, an order-identity gap, and a permission gate that silently skipped half of all chats. Each one was found, proven, and fixed before it became a headline.
The method is fixed: drafts-only first, test on a copy, instrument from day one, white-label always. The background, briefly: Boston College, time inside the European Parliament, production AI at a European startup. The rest is the work above.
Lines below. Replies are quick and written. For NDA work, email is the right door.