Luma AI
Research Scientist / Engineer – Reinforcement Learning Infrastructure
$195K–$395K
Redwood City, CA +3 · Hybrid · Full-time · Posted Jul 24, 2026
- Location
- Redwood City, CA
Remote, US
New York, NY
Seattle, WA - Experience
- Hands-on RL post-training at meaningful scale is required; no number of years stated.
About the job
Build and scale distributed reinforcement-learning post-training infrastructure for Luma’s models, spanning training, rollout generation, agent environments, rewards, and evaluation.
What you'll do
- Design and scale distributed RL post-training workloads across thousands of GPUs.
- Build high-throughput rollout systems, agentic environments, and reward infrastructure.
- Develop tools for evaluating, monitoring, debugging, and improving the stability and efficiency of RL runs.
What they ask for
- Hands-on experience post-training LLMs with RL (PPO/GRPO-family, RLHF, RLVR) at meaningful scale.
- Extensive distributed PyTorch training and parallelism for foundation models.
- Experience building RL environments, reward functions, verifiers, or evaluation harnesses for LLM agents.
- Deep familiarity with RL post-training frameworks, rollout inference engines, and GPU-cluster networking and communication libraries.
Nice to have
- Running RL training across 100+ GPUs, including asynchronous or disaggregated trainer/rollout architectures.
- Containerization and orchestration for large environment fleets and sandboxed workloads; research or open-source contributions in RL for LLMs.
Pay
$195K – $395K (base salary)
Tools
PyTorch, FSDP, Tensor/Pipeline/Expert Parallel, veRL, OpenRLHF, TRL, Ray, vLLM
AI focus
Reinforcement-learning post-training for LLMs, including agentic environments, reward and verification systems, and training efficiency and stability.
Model-written summary · the original posting is authoritative