I am a Staff Research Scientist and Research Lead at Meta Superintelligence Labs. My current research focuses on:
1. adaptive, steerable, and contextual-aware alignment methods
2. reinforcement learning algorithms
3. (multi) agentic modeling
4. new model architectures
In the past, I also worked on Federated Learning, optimization algorithms, and efficient on-device deep learning models.
E-mail: my firstname dot lastname At *mail.com, where * = g
See more in my google scholar page.
Our new collaborative alignment method is discussed in the "The most interesting thing in tech" podcast by Nicholas Thompson (CEO @ The Atlantic).
The algorithm I developed for traffic time-series forecasting is used by the California Department of Transportation (CalTran) in the Connected Corridor project. Media coverages: 1, 2, 3
"Personalized Federated Learning for Assistant Systems". U.S. Patent Application No. 16/815,990.
"Active Federated Learning for Assistant Systems." U.S. Patent Application No. 16/815,960.
