Research Scientist - RL Training

Role overview Advance reinforcement learning methods for training and aligning large language models, with a particular focus on the data and reward signals that help produce reliable, generalizable behavior. This research role combines experimentation with practical delivery: turning research into preference datasets, reward models, and training-ready corpora that can support model development at scale.

Responsibilities - Investigate and implement approaches such as RLHF, RLAIF, GRPO, DPO, and reward modeling, translating them into useful training data and signals. - Design data pipelines for producing and curating high-quality training signals, including AI-assisted annotation workflows. - Prototype and refine end-to-end reinforcement learning training recipes and assess how they affect the resulting data products. - Collaborate with research scientists, machine learning engineers, and delivery teams to turn research findings into usable solutions. - Track progress in alignment research, distributed LLM training, and scalable reinforcement learning, and assess relevant methods for practical use. - Contribute to research publications and internal technical documentation.

Requirements - Deep expertise in reinforcement learning from human or AI feedback, reward modeling, and credit attribution, including an understanding of the role data plays. - Experience training or fine-tuning large language models at scale, ideally models with 30B or more parameters, and familiarity with distributed training. - Strong Python skills and experience with PyTorch and Hugging Face, plus hands-on use of reinforcement learning frameworks such as Verl or SkyRL. - Software engineering skills sufficient to build research prototypes that others can run, extend, and integrate into production workflows. - Comfort working through open research questions in a fast-iterating environment with changing technical needs. - A PhD in machine learning, reinforcement learning, or a related discipline is strongly preferred; substantial industry experience may also qualify.

Nice to have - Experience with machine learning infrastructure and cloud tools such as AWS, GCP, Kubernetes, or Slurm. - Experience operating large-scale reinforcement learning training pipelines.

Benefits and work setup The role offers scope to influence research priorities and technical direction, with opportunities to deepen expertise, develop leadership skills, and learn across functions. The stated salary range is $200,000–$350,000 USD; actual compensation depends on skills, qualifications, experience, and location.