Responsibilities:

• Designing and implementing RL pipelines from reward modeling to policy optimization
• Optimizing RL training stability and sample efficiency for large models
• Verifying numerical correctness across inference and training
• Performance engineering on trainer-inference communication
• Validating methods from recent publications

Qualifications:

• Hands-on experience with reinforcement learning in production systems
• Deep understanding of policy-space methods (GRPO, PPO, etc.)
• Experience profiling distributed systems

Preferred:

• History of OSS contributions
• Knowledge of TorchTitan and SGLang or vLLM