A personal AI assistant that gets real things done
The problem
Most AI assistants can chat, but few can act, reliably completing multi-step tasks like bookings, calls, and admin over the channels people already live in: iMessage and WhatsApp. Trust required more than a good demo; it required proof the assistant wouldn't fail silently in the real world.
My role
As founding PM, I own product vision, GTM positioning, and the prioritized roadmap end-to-end: from what the assistant should be trusted to do first, to how we prove it's ready for wider release.
What I built
An AI evaluation framework spanning user-journey, channel, feature, and risk-level dimensions, run through an automated test runner integrated directly into iMessage. It lets us systematically evaluate assistant reliability before anything reaches a real user, catching failure modes a manual spot-check never would.