We are particularly interested in candidates with substantial experience in one or more of the directions below. You do not need expertise in every area. We value clear research questions, careful experimentation, strong implementation skills, and a willingness to collaborate.
Vision-Language Models & Multimodal Reasoning
Grounding language in visual evidence; learning and evaluating reasoning across images, video, text, and other modalities.
Relevant background: Experience training or adapting VLMs, designing multimodal evaluations, or studying grounded reasoning.
Video Understanding & Reasoning
Understanding events, temporal relationships, and long-horizon context across clips, viewpoints, and videos.
Relevant background: Experience with video-language models, temporal grounding, long-video reasoning, or efficient video representations.
Video Generation & World Modeling
Controllable, temporally coherent generation and predictive representations that connect visual understanding with planning.
Relevant background: Experience with video diffusion models, generative modeling, motion representations, or learned dynamics.
3D Human & Scene Reconstruction
Recovering human shape, motion, and dynamic scenes from visual observations, including Gaussian splatting and neural rendering.
Relevant background: Experience with human mesh recovery, multi-view geometry, 3D/4D reconstruction, or differentiable rendering.
Physical AI & Robotics
Connecting multimodal perception and reasoning to robot planning, actions, and feedback from the environment.
Relevant background: Experience with robot learning, vision-language-action models, reinforcement learning, or simulation-based evaluation.
Agentic Systems & Harness Optimization
Improving the tools, memory, context management, orchestration, and feedback loops around AI models to make multi-step tasks more reliable and efficient.
Relevant background: Experience building and evaluating tool-using agents, execution environments, or systematic agent-improvement pipelines.