I’m drawn to the point where AI stops being a model in isolation and becomes part of everyday life. Much of my work at Meta has lived there: real-time voice and multimodal systems, private memory, and agent infrastructure for wearable devices. These systems have to disappear into the experience—fast enough to feel natural, dependable at scale, and private by construction.
I tend to work across the boundaries that make that possible: from product behavior and system architecture to low-latency services, encrypted retrieval, and confidential computing with hardware attestation. What interests me is not any one layer, but the act of bringing them together into something coherent.
Lately, I’ve been thinking about what comes after the assistant: systems that remain present, learn through interaction, and build useful context over time without taking ownership away from the person using them. That leads me toward agents and memory, real-time multimodal intelligence, privacy-preserving infrastructure, and reinforcement learning.