AI Research Scientist, Infrastructure Engineer, Reinforcement Learning

What you'll do

• Design and implement distributed RL training stacks using data/pipeline parallelism integrated with AMD schedulers.
• Build high-throughput rollout workers, trajectory stores, and reward computation pipelines with versioning.
• Instrument jobs to detect and debug issues like NaNs, stragglers, and Out-of-Memory errors.
• Implement autoscaling and preemption-safe checkpointing for large GPU fleets.
• Develop researcher-facing APIs and experiment templates to improve scientist productivity.
• Manage infrastructure reliability through on-call rotations, runbooks, and postmortems for training incidents.
• Optimize performance through C++/Python tuning and I/O optimization for containerized workloads.

What we're looking for

• A Bachelor's degree in Computer Science is required.
• A Master's or PhD in Computer Science is preferred.
• Experience with PyTorch, JAX, and NCCL/MPI-style distributed training is required.
• Expertise in GPU cluster orchestration and large-scale experiment management is required.
• Proficiency in C++ and Python performance tuning and I/O optimization is required.
• Experience building high-throughput rollout workers and reward computation pipelines is preferred.
• Experience with LLM post-training pipelines or RL training infrastructure is preferred.
• Demonstrated track record in machine learning platforms with deep systems expertise is preferred.