Training agents that perceive and act. Research Scientist, Meta Superintelligence Labs · 2025–
I train multimodal models to act in the real world. Start from the capability we want, build the tasks, environments, rewards, and synthetic data that teach it, then train with RL.
At 
Meta Superintelligence Labs I lead synthetic data and RL for Muse, a real-time omni model for personal agents, and built the offline RL system behind Vibes. Before that, I worked on Visual Intelligence at 
Apple and foundation models at 
MIT-IBM Watson. I got my Ph.D. in Computer Science at 
MIT, and B.A. in Physics at 
UC Berkeley. Outside work I read, write, bike, and hike.
Muse Realtime Voice carries the conversation as a stream of speech tokens, and Muse Realtime Avatar turns the same stream into a live, expressive character for as long as the conversation runs. I lead synthetic data and RL, and work on the distillation that makes it fast enough to serve in real time.
Built the agentic long-form video workflow and the offline RL system behind Vibes, Meta AI's personalized video feed, and led its human-evaluation data.
Built the datasets for on-device visual question answering on iPhone with privacy-preserving VLMs, and post-trained Apple's foundation model for proactive question answering on egocentric video. Shipped in Apple Intelligence at WWDC 2025.
Given a target capability, synthesize the tasks, the environment, and the reward, then train with RL. No labeled dataset required; the task definition is the input.
JetMoE-8B, an open-source mixture-of-experts model trained for a fraction of the usual cost that outperforms Llama2-13B, with a synthetic corpus for pre-training and post-training.
Efficient on-device task planning and decomposition, trained on a synthetic planning dataset with a multi-LoRA post-training pipeline.
At COLM 2026 in San Francisco, Oct 6–9. Email me to meet at the poster sessions or the Lifelong Agents workshop.
Read more →Muse Realtime Voice and Avatar announced at the Meta Connect 2026 keynote: one streaming system for conversation, speech, and a live, expressive avatar. I lead synthetic data and RL for Muse.
Read more →Muse Spark 1.1 released, a multimodal reasoning model for agentic tasks, alongside the Meta Model API public preview.
Read more →When the headline metric becomes output per joule, a technology has become a utility. Intelligence just did, and the money will go where it always goes: upstairs.
Machines calculate. Humans commit. The gap is not intelligence — it is what you are willing to wager once the evidence runs out.
Org-chart agent systems import human coordination limits into machines that don't have them. Let the task emit the graph instead.
