About Me

I am a Senior Research Scientist at Meta Superintelligence Labs (MSL) where I currently focus on multimodal (speech) LLM post-training, with an emphasis on reward modeling for both verifiable and non-verifiable domains, and reinforcement learning. I also work on neural codec training and diffusion modeling for speech synthesis from Meta’s multimodal models.

I was previously a PhD student in the EECS department at MIT advised by Professor Gregory Wornell. I defended my Ph.D. thesis titled “Score Estimation for Generative Modeling” in May 2025. Before that I completed my M.S. degree in EECS from MIT in January 2022 and my B.S. degree in Electrical Engineering from the University of Illinois at Urbana-Champaign in 2019.

Research Interests

• Reward Modeling & LLM Post-training: Training reward models and judges to enhance multimodal LLM outputs using RLHF and DPO (Generative Speech Reward Model, MSL ‘26).
• Efficient Generative Modeling: Designing score estimation, flow matching, and distribution matching objectives that cut diffusion sampling from hundreds of network evaluations down to a single step (Score-of-Mixture Training, ICML ‘25 Spotlight), and that repurpose pretrained diffusion priors as solvers for inverse problems such as source separation (α-RGS, NeurIPS ‘23).
• Neural Data Compression: Using generative models as priors for low-bitrate compression and tokenization (Efficient Video Compression Transformers, Google Research ‘23 and Model Code Separation, MIT ‘24).