I am an AI research scientist at Meta on the Multimodal Recommendations team, where we look for ways to use models to understand content better with the goal of improving search/ranking results across Meta. I work on topics including (1) developing LLM reasoning models for trust and safety enforcement, and (2) improving the efficiency and quality of text extraction models (OCR/ASR). This work has led to robust improvements in key business metrics like time spent on Instagram recommendations.

Outside of my work at Meta, I led the development of POPri, which used RL (reinforcement learning) to achieve a step-change improvement in synthetic data generation under privacy constraints. I was invited to give a talk on POPri at OpenAI. Before Meta, I obtained my Ph.D. from CMU advised by Prof. Giulia Fanti, where among other work, I co-led SquirRL which pioneered the use of RL for identifying security weaknesses in blockchain protocols.

Before my Ph.D., I was an undergrad at Princeton University.

Twitter / CV / Google scholar / Email / Linkedin