Figure
Helix AI Engineer, Video Pretraining
About this role
Figure AI is seeking a Helix AI Engineer to develop large-scale video foundation models that learn from raw video data to enable autonomous humanoid robots. The role involves designing pretraining strategies, building efficient data pipelines, and collaborating with cross-functional teams to integrate models into the robotics autonomy stack. This position requires 5 days/week in-office presence in San Jose, CA.
What you'll do
- Design and train large-scale video foundation models on diverse internet and robot-collected datasets
- Develop pretraining strategies that capture temporal dynamics and motion interactions from video sequences
- Build models with transferable representations for perception, tracking, prediction, and control tasks
- Implement efficient data pipelines and distributed training systems for high-throughput video processing
- Optimize model performance across compute, memory, and training efficiency constraints
- Design evaluation frameworks and benchmarks for measuring temporal understanding and generalization
What they're looking for
- Large-scale video model training and sequential data handling
- Deep learning architectures for video and vision systems
- PyTorch and distributed training frameworks
- Large-scale model pretraining and dataset curation
- GPU cluster and distributed systems experience
- Software engineering and scalable systems design
- Experimental rigor and rapid iteration
- Video diffusion, autoregressive modeling, or world models (bonus)
Opens the application — the Jobs AI extension fills it for you. Set up autofill
Opens the official application on the employer’s site. No login required.
Figure
Figure develops advanced humanoid robots powered by AI technology. The company is hiring engineers across mechanical design, firmware development, manufacturing, quality assurance, and security to build and refine its autonomous robotic systems.
View all jobs at FigureLikely interview questions
- Walk us through a large-scale video pretraining project you've led—what dataset did you use, what architecture choices did you make, and how did you measure success?
- How do you approach designing pretraining strategies that capture temporal dynamics and motion from raw video? What trade-offs have you encountered?