Applied Scientist

About the Role

Applied Scientists own the mastery of new frontier model capabilities. This includes developing post-training recipes for learning a wide range of dexterous tasks - from pushing the frontiers of science lab automation, to generalizing manufacturing and warehouse logistics, to designing open-ended evaluations that challenge what the models can do, and more. You are responsible for hill-climbing broad sets of manipulation capabilities until they are robust, general, and reusable across multiple applications. This role is a hybrid of research, engineering, and operations, and we expect excellence in all three areas.

This is an empirical, model science-first role. You will define what "good" looks like for a capability, build the evaluations that measure it, run experiments across model variants and data recipes, analyze the results, and drive the post-training loop. You will leverage post-training, data, task design, and rigorous experimentation, to compound model capabilities: anything mastered for one application can become a strength of the foundation model for entire industries.

You'll Be Responsible for

• Owning end-to-end improvement of target capabilities, from baseline performance to mastery
• Designing task suites and evaluation protocols that faithfully capture what research, science partners, and customers need, and that generalize beyond any single deployment
• Coordinating data collection, training and fine-tuning models, sweeping data mixtures and configurations, and orchestrating large-scale evaluations
• Running rigorous experimental analysis across model variants - isolating variables, quantifying statistical confidence, and diagnosing failure modes from rollout data
• Using the stack in new ways rather than redesigning it - and when a capability is fundamentally restricted, you'll work with the pretraining and post-training teams to improve the model

You Might Thrive in This Role If You

• Are comfortable with data analysis at scale: slicing roll out logs, computing metrics, visualizing results, and investigating what happened and why
• Care deeply about evaluation, benchmarking, and failure analysis - and are continuously diligent in the face of seemingly repetitive, but subtly changing task evaluations
• Enjoy rapid iteration with real-world feedback loops, and are willing to get hands-on with robots and physical task setups when the experiment demands it
• Are motivated by making capabilities work for real applications
• Can communicate findings clearly to both research teams and partner-facing teams
• Have a strong background in machine learning and/or robot learning (e.g. a PhD in a relevant area)