Applied Scientist

Generalist • San Francisco, CA

21h

About The Position

Generalist is building embodied foundation models, initially focused on dexterity, to help robots intelligently interact with the physical world. The company combines large-scale AI and robotics, with a team of researchers, roboticists, and company builders from frontier labs and organizations.

Applied Scientists develop post-training recipes and advance frontier model capabilities across dexterous tasks, including science lab automation, manufacturing, and warehouse logistics. They improve broad manipulation capabilities until they are robust, general, and reusable across applications. The role combines research, engineering, and operations, with an empirical, model-science-first approach. Scientists define capability goals, build evaluations, run experiments across model variants and data recipes, analyze results, and drive the post-training loop.

Requirements

• Strong background in machine learning and/or robot learning.
• Ability to analyze data at scale, including slicing rollout logs, computing metrics, visualizing results, and investigating outcomes.
• Care about evaluation, benchmarking, and failure analysis, with diligence across repetitive but subtly changing task evaluations.
• Enjoy rapid iteration with real-world feedback and willingness to work hands-on with robots and physical task setups when needed.
• Motivation to make capabilities work for real applications.
• Ability to communicate findings clearly to research teams and partner-facing teams.

Nice To Haves

• A PhD in a relevant area.

Responsibilities

• Own end-to-end improvement of target capabilities, from baseline measurement to demonstrated mastery.
• Design task suites and evaluation protocols that capture the needs of research, science partners, and customers, and generalize beyond individual deployments.
• Coordinate data collection, model training and fine-tuning, sweeps of data mixtures and configurations, and large-scale evaluations.
• Run rigorous experimental analyses across model variants by isolating variables, quantifying statistical confidence, and diagnosing failure modes from rollout data.
• Develop post-training recipes for dexterous tasks and improve manipulation capabilities to make them robust, general, and reusable.
• Define capability goals, build evaluations, run experiments, analyze results, and drive the post-training loop.

What This Job Offers

• Job Type: Full-time
• Education Level: No Education Listed