AI Research Scientist - Infrastructure Engineer, Reinforcement Learning

About the Role

We are hiring an AI Research Scientist - Infrastructure Engineer, Reinforcement Learning, to own reinforcement learning infrastructure at scale—including distributed policy and value training, rollout generation, logging, checkpointing, and researcher-facing APIs across large GPU fleets. You make RL scientists productive by improving throughput, fault tolerance, reproducibility, and observability—turning fragile notebooks into reliable systems that RSI, generalizing-HW, and RL research programs depend on.

Responsibilities

• Design and implement distributed RL training stacks (data parallel, pipeline parallel, or hybrid) integrated with AMD’s schedulers and storage
• Build high-throughput rollout workers, trajectory stores, and reward computation pipelines with versioning and audit trails
• Instrument jobs for debugging (NaNs, stragglers, OOMs), implement autoscaling and preemption-safe checkpointing
• Collaborate with research scientists on experiment templates, hyperparameter sweeps, and safe promotion paths from research to wider team use
• Drive reliability: on-call rotations, runbooks, and postmortems for infra incidents affecting RL training

Qualifications

• Strong systems track record in machine learning (ML) platforms with deep systems expertise and demonstrated technical impact
• Deep experience with PyTorch (or JAX), NCCL/MPI-style distributed training, and GPU cluster orchestration
• Prior ownership of RL training infra, LLM post-training pipelines, or large-scale experiment management
• Proficiency in C++/Python performance tuning, I/O optimization, and containerized workloads

Academic Credentials

Bachelor’s degree required; Master’s or PhD preferred in Computer Science for research-heavy collaboration depth.

Benefits

AMD offers a comprehensive benefits package. For more details, see AMD benefits at a glance.