Member of Technical Staff, Model Training (post-training research)

The work

Investigate how learning algorithms, training data and evaluation interact to improve reasoning, generalization and agent abilities. Form research hypotheses, run ambitious experiments and turn evidence into scalable methods for stronger language and multimodal models.

• Implement, research and scale supervised fine tuning, reinforcement learning and on policy distillation methods.
• Study objectives, optimization behavior, exploration and credit assignment to improve stability, sample efficiency and acquired capabilities.
• Refine training recipes by coordinating data mixtures, sampling, curricula and evaluation design.
• Develop learning methods based on expert demonstrations, preferences, corrections and generated trajectories, identifying where human input matters most.
• Investigate how generators, reward models and graders evolve together and improve both performance and training signal quality.
• Design controlled experiments, ablations and evaluations of reasoning, generalization and realistic downstream performance.
• Own research from promising initial questions and algorithm implementation through large experiments and validation.

Experience and skills

• Strong grounding in ML, deep learning, optimization and contemporary training approaches.
• Practical experience training, fine tuning or post training large language or multimodal systems.
• Deep expertise in at least one of reinforcement learning, distillation, reward modeling, data selection or learning dynamics.
• Strong Python and PyTorch skills for implementing, diagnosing and optimizing complex training pipelines.
• Experience designing controlled experiments, interpreting behavior and converting findings into research directions.
• Independent judgment on ambiguous questions, hypothesis formation and rapid evidence based iteration.
• A research mindset emphasizing scientific rigor, repeatability and measurable progress.
• Helpful experience includes RLHF, RLAIF, preference optimization and reinforcement learning for reasoning or agents.
• Distributed training, GPU optimization and large experiments are valuable additions.
• Reward model, verifier, grader or automated evaluation development is welcome.
• Research publications, open source contributions or production work in post training or RL are advantages, as are synthetic data, curriculum learning and iterative improvement studies.

Working arrangements

This full time position is remote, with national base pay of $200,000 to $320,000. All employees are eligible for equity; performance bonuses may also apply according to the role and policies.

Benefits include up to 100% reimbursement of health insurance premiums, paid time off and a 401(k) with an employer match.