Featured Job

Research Scientist

San Francisco, California Full-time On-site $200k — $335k per year 11/18/2026 Job ID: 000597
Apply Now
Reinforcement Learning Policy Gradient Methods Variance Reduction On-Policy Training Off-Policy Training

Summary

What you’ll impact

The Senior Reinforcement Learning Research Scientist designs, implements, and scales cutting‑edge RL algorithms and training pipelines. They work with large GPU clusters and distributed frameworks like PyTorch, JAX, and DeepSpeed, leveraging low‑precision numerics and system optimizations. Key tasks involve running RL experiments, developing post‑training methods, diagnosing instability or reward‑hacking, and enhancing inference efficiency. They partner with applied researchers to move advances into production and contribute to open‑source ML infrastructure.

Responsibilities

What you'll do

  • Design, run, and analyze RL experiments, scoping them carefully to separate real algorithmic gains from noise
  • Develop and improve post-training algorithms, reasoning about why they work (or don’t) and diagnosing failures like instability, reward hacking, and collapsing exploration
  • Design and optimize our RL training and inference pipelines across large GPU clusters
  • Understand how systems choices affect learning (low-precision numerics, off-policy staleness, distributed training edge cases, etc.)
  • Partner with applied researchers to bring frontier post-training capabilities into production deployments

Requirements

What you’ll bring

  • A strong foundation in probability, statistics, machine learning, and reinforcement learning, including policy gradient methods, variance reduction, and the tradeoffs between on-policy and off-policy training
  • The ability to learn quickly and navigate ambiguous or underspecified problem settings
  • Fearlessness and curiosity to understand things, with an eye for details in specific algorithms or the training stack
  • Experience programming with and managing training jobs on large-scale GPU systems
  • A bias toward fast implementation, paired with a high bar for rigor, reliability, and efficiency
  • Familiarity with open-weights models (architecture and inference)
  • Background in reinforcement learning or integration of inference with RL training loops
  • Published research or substantial projects in RL, post-training, or LLM reasoning and alignment
  • Demonstrated technical creativity through published projects, OSS contributions, or side projects
  • Experience with distributed training frameworks (PyTorch, JAX, DeepSpeed)
  • Background in high-performance computing or working with large-scale clusters
  • Contributions to open-source ML infrastructure

Ready to Move Forward?

Apply now and our recruiting team will reach out with next steps, interview guidance, and client insights tailored to this role.