One line

Gavin Guo is a research scientist at Meta Superintelligence Labs who trains real-time multimodal agents.

Short

Gavin Guo is a research scientist at Meta Superintelligence Labs, where he trains real-time multimodal agents, co-leads post-training, and leads synthetic data for Muse Realtime Avatar. He previously worked on Visual Intelligence at Apple, shipped in Apple Intelligence at WWDC 2025, and on foundation models at the MIT-IBM Watson AI Lab. He holds a Ph.D. in Computer Science from MIT and a B.A. in Physics from UC Berkeley.

Long

Gavin Guo is a research scientist at Meta Superintelligence Labs. He trains real-time multimodal agents that talk, listen, and act while the interaction is still going. He co-leads post-training and leads synthetic data for Muse Realtime Avatar, announced at Meta Connect 2026, and developed offline RL for Vibes, Meta AI’s personalized video feed. Before Meta, he was an AIML Resident at Apple, where he built the datasets for on-device visual question answering and post-trained Apple’s foundation model for Visual Intelligence, shipped in Apple Intelligence at WWDC 2025. Earlier he worked on foundation models at the MIT-IBM Watson AI Lab, co-led JetMoE, an open mixture-of-experts model whose chat model beats Llama2-13B-Chat on MT-Bench, and co-authored Synthetic Data RL and Octo-planner. He holds a Ph.D. in Computer Science from MIT and a B.A. in Physics with Highest Honors from UC Berkeley. His current interests are how agents keep state and act continuously, and how to turn their interactions and failures into training signal.

Headshot

Download headshot (JPEG, 720×720)

Links

Google Scholar · GitHub · LinkedIn · X · Email