Senior RL Post-Training Systems Engineer

NVIDIA; Massachusetts; On-site; USD 224,000 - 357,000; Full time

Job summary

NVIDIA is building an RL Frameworks engineering team to develop open-source tools and infrastructure for AI researchers and post-training teams. The role spans the full software stack, from researchers to distributed runtimes, including VeRL, Miles, TorchTitan, Ray, Monarch, and more.

You will architect and build RL post-training infrastructure that scales from single-GPU experiments to production across thousands of nodes, tuning loops for peak performance while improving framework usability

Qualifications

• MS or PhD in Computer Science, Computer Engineering, or related field (or equivalent).
• 5+ years in distributed systems, HPC, or ML systems engineering.
• Strong Python and C/C++ proficiency.
• Experience building large-scale distributed systems in production.
• Excellent verbal and written communication across teams.

Responsibilities

• Architect and build RL post-training infrastructure that scales from single-GPU to thousands of nodes.
• Tune training-inference-rollout loops on GPUs, CPUs, and LPUs for performance.
• Improve open-source RL frameworks and partner with hardware and runtime teams.
• Ensure fault tolerance, elastic scaling, and fast restarts for long-running jobs.
• Collaborate with researchers and partners to prioritize RL workload capabilities.

Skills

Python

C/C++

Distributed systems

Communication

Research experience

Education

MS or PhD in CS/CE or equivalent

Tools

PyTorch

Kubernetes

Ray

Monarch

Job description

NVIDIA is building an RL Frameworks engineering team to develop open-source tools and infrastructure for AI researchers and post-training teams. The role spans the full software stack, from researchers to distributed runtimes, including VeRL, Miles, TorchTitan, Ray, Monarch, and more.

You will architect and build RL post-training infrastructure that scales from single-GPU experiments to production across thousands of nodes, tuning loops for peak performance while improving framework usability