About me

I am a research scientist at Luma AI, working on multimodal reasoning and generation. Previously, I was a research scientist at NVIDIA Research, where I was a core contributor to NVIDIA Cosmos. I obtained my Ph.D. in 2024 from the University of Illinois Urbana-Champaign (UIUC), and my B.S. in 2019 from UIUC with double majors in Physics and Statistics & Computer Science.

My research focuses on post-training for multimodal models and LLMs, including reasoning, RLHF, reward modeling, and synthetic data curation.

Selected Projects

Luma Uni-1 — A frontier unified understanding and generation model.

Core contributor to Uni-1
#1 Elo rating in human preference evaluation across leading image generation models

Uni-1 Elo Chart

NVIDIA Cosmos — Open models for physical AI.

Core contributor to Cosmos-Reason1 (2M+ total model downloads) and Cosmos-Reason2, reasoning vision-language models for Physical AI
Core contributor of Cosmos World Foundation Models (8K+ GitHub Stars)

RLHFlow — Open-source RLHF for LLM post-training.

Developed ArmoRM, a multi-objective reward model that achieved #1 on RewardBench (May 2024), surpassing GPT-4 judges; 400K+ downloads on HuggingFace
Built the first open-source online iterative DPO framework