Senior Manager, Software Engineering - RL Post-Training Frameworks

NVIDIA•Us, CA

13d•Hybrid

About The Position

Can you bring together globally distributed teams and the systems they build into a production-quality reinforcement learning ecosystem for researchers and model builders? Reinforcement learning post-training is where modern AI systems learn to reason, use tools, follow detailed instructions, and act as agents. Making that capability work at scale creates one of the most demanding systems problems in AI: a single RL run ties together inference, rollout, reward and critic evaluation, and training. At frontier scale, these loops have to run reliably across GPUs, CPUs, networking, storage, and open-source runtimes. You will lead the work to build, extend, and harden the rapidly evolving pieces to compose cleanly and scale with the most ambitious RL projects on NVIDIA's platforms. To meet that challenge, NVIDIA is building an RL Frameworks engineering team for the open-source tools and infrastructure that researchers, model builders, and external partners depend on. We are looking for a Senior Software Engineering Manager to set strategy, build the team, and convert emerging technical, customer, and partner signals into clear engineering priorities. The role spans RL frameworks such as VeRL, Miles, Slime, SkyRL, TorchTitan, and related post-training stacks, along with the systems those stacks build on and compose with: Megatron-Core, Ray, Monarch, NIXL, SGLang, Kubernetes, and NVIDIA platform libraries. Come build the ecosystem that the next generation of AI will rely on!

Requirements

MS or PhD in Computer Science, Computer Engineering, or a related field (or equivalent experience)
10+ years of software engineering experience in distributed systems, AI frameworks, ML infrastructure, high-performance computing, or systems software
4+ years as an engineering manager for software teams
Strong technical background in distributed AI systems, including the ability to reason across training, inference, orchestration, and end-to-end performance, and challenge architecture and performance tradeoffs with senior engineers
Experience defining domain-level technical strategy, making build-vs-buy or upstream-vs-internal investment decisions, and creating multi-team execution plans
Ability to drive engineering work across organizational boundaries, influence without direct authority, and communicate tradeoffs clearly to senior leaders and executives
Experience hiring and leading engineering teams, developing technical leaders or new managers, and creating staffing plans for constantly evolving technical domains
Experience establishing workflows, success criteria, metrics, or decision gates that improve engineering execution across teams
Background collaborating with open-source communities, research teams, external partners, or customer-facing teams

Nice To Haves

Hands-on experience with RL post-training frameworks or algorithms such as RLHF, PPO, GRPO, DPO, reward modeling, VeRL, Miles, Slime, SkyRL, OpenRLHF, NeMo-Aligner, or TorchTitan
Background with runtime and orchestration systems such as Ray, Monarch, Kubernetes, Slurm, or comparable actor- and task-based systems
Experience scaling workloads across thousands of GPUs or heterogeneous systems, including fault tolerance, elastic recovery, stragglers, resource contention, or benchmark reproducibility
Familiarity with NVIDIA platform components such as CUDA, NCCL, cuDNN, TensorRT-LLM, Transformer Engine, Nsight, NeMo, or Megatron-Core
Demonstrated ability to turn customer or partner needs into reusable upstream improvements rather than one-off support

Responsibilities

Own NVIDIA's RL post-training frameworks strategy: where we invest directly, where we partner upstream, and how we prioritize based on customer impact, ecosystem leverage, technical feasibility, and opportunity cost.
Use systems depth to evaluate architecture and performance claims across training, inference, rollout, orchestration, and the NVIDIA platform.
Help expert teams converge on integrations that improve RL framework quality and user value, then turn those decisions into measurable execution plans.
Benchmark and establish reproducibility criteria.
Deliver across open-source frameworks and distributed runtimes.
Partner closely with product management, research, DevRel, customer-facing teams, hardware, CUDA, networking, math libraries, compilers, and external open-source collaborators.
Build the team: recruiting and developing managers and senior ICs, creating an effective US/APAC operating model, reviewing capacity against commitments, and setting clear ownership and decision rights.
Coach engineers to contribute credibly in open-source ecosystems and carry NVIDIA's priorities through high-quality upstream work.
Turn open technical and partner questions into concrete and measurable action, set delivery goals, and hold the quality bar.
Ensure validated, valuable work rather than work that merely lands, plus durable open-source improvements that make RL workloads run well on NVIDIA systems.