Work
Selected work across proactive coaching, agentic companions, and SDK surfaces embedded inside real products.
BodyBuddy
Most health apps wait for the user to log something. BodyBuddy acts more like a coach: it checks in, remembers context, notices patterns, and keeps the thread going.
Product direction, AI coaching behavior, onboarding, retention loops, and the shipped iMessage-style product experience.
Privy
Privy's components were a customer-facing integration layer, not a marketing page. Any change had to respect live sites, API expectations, and teams already building on top of it.
Re-architected the core component system with clearer APIs, safer composition patterns, and a migration path for a surface already embedded across thousands of customer sites.
This is the kind of product engineering AI work often depends on: the interface has to be powerful enough for builders, stable enough for production, and simple enough to adopt.
Treasure
A useful AI companion cannot just chat. It needs a clear job, a charming product surface, and enough user control to feel helpful instead of chaotic.
AI companion experiences for Treasure, including Smol Clyde: a desktop assistant designed around projects, folders, tools, and user-controlled tasks.
A consumer-facing agentic workflow where the personality, permission model, task loop, and interface all had to work together.