I’m a Member of Technical Staff at Thinking Machines Lab, working on Inkling with a focus on multimodal capabilities.
Previously, I worked on visual generation and world modeling at Luma AI and NVIDIA Research.
I earned my Ph.D. from UIUC, where I studied machine learning, including out-of-distribution generalization and RLHF. My past research also includes reward modeling.
Thinking Machines Lab 2026
Multimodal reasoning.
Open weights.
I helped ship Inkling, an open-weights foundation model with native reasoning across text, images, and audio, controllable thinking effort, and fine-tuning on Tinker.
NVIDIA Research
World models for physical AI.
Luma AI
Core contributor to Uni-1, a unified model for visual understanding and image generation.
Open-source research
Open-source tools for LLM post-training.
Reasoning, reinforcement learning, and multimodal generation.
Reinforcement learning for diffusion models
LLM reasoning and post-training
NVIDIA · World foundation models for physical AI
ArmoRM · Multi-objective reward modeling
Reward modeling and online alignment
Personalized alignment and preference optimization
Helped ship Inkling, Thinking Machines Lab’s first open-weights model.
Joined Thinking Machines Lab.
Luma Uni-1 is released.
Cosmos-Reason1 is released.
Cosmos World Foundation Model is released.
ArmoRM accepted to EMNLP; RLHF Workflow accepted to TMLR.
I joined Thinking Machines Lab in June 2026. Before that, I was a Research Scientist at Luma AI and NVIDIA Research.
I earned my Ph.D. in Electrical & Computer Engineering from the University of Illinois Urbana-Champaign in 2024, and my B.S. from UIUC in 2019, with double majors in Physics and Statistics & Computer Science.
haoxiang096@gmail.com