Job Details

Our client is one of the most exciting companies in robotics right now. Founded by a repeat unicorn founder with multibillion dollar exits, they've raised over half a billion dollars to pursue a single audacious mission: a helpful robot in every home. Their lean, elite team of ~70 comes from Tesla, OpenAI, Google, Cruise, and Pixar. The best in the world, building something the world has never seen.

The Role

You'll own the full stack of a unified multimodal foundation model, from pretraining to deployment on real robotic hardware. This is foundational research with direct physical impact. No hand-offs, no bureaucracy, just hard problems and the resources to solve them.

What You'll Do

• Design and train large-scale multimodal architectures where vision, language, and kinematics share a unified representation
• Research and implement genuine cross-modal reasoning, not shallow associations
• Run end-to-end training loops: data curation, experiment design, failure diagnosis, and iteration
• Deploy models directly onto robotic hardware and optimize for edge inference

What You Bring

• Strong Python, C++, or Rust
• A proven track record training and deploying large-scale multimodal models in production
• Deep fluency in LLM pretraining, post-training, and RL at scale
• Comfort owning very large GPU cluster experiments from start to finish

Why This

The problems are unsolved. The team is world-class. The funding is exceptional. If you want to do the most consequential work in embodied AI right now, with the people and resources to actually ship it, this is the role.