Research Scientist / Engineer – Reinforcement Learning Infrastructure

✓ Verified Job

Canonical indexed version, validated from employer's careers page.

Source: Company careers pageValidated by: Fursa AI

Last checked: 2 hours agoStatus: Live

Job Summary

Build and scale distributed reinforcement learning post-training infrastructure for large multimodal foundation models. You’ll orchestrate trainer/rollout/environment/reward workloads across thousands of GPUs, optimize high-throughput rollout generation with inference engines (vLLM, SGLang), and design hermetic RL environments. Own reward, evaluation, monitoring, and debugging tooling to keep runs stable and diagnose regressions, while collaborating with researchers to productionize new RL approaches.

Location: Redwood City, London

Workplace: Hybrid

Employment Type: Full time

Job Function: Data Science & Machine Learning

Key Responsibilities

•Design, build, and scale distributed RL post-training systems for large multimodal models, orchestrating trainer, rollout, environment, and reward workloads across thousands of GPUs.
•Build and optimize high-throughput rollout generation by integrating inference engines (vLLM, SGLang) into the training loop and applying weight synchronization and asynchronous/off-policy schemes.
•Design and implement RL environments for agentic, multi-step tasks with sandboxed code execution, tool/computer use, and multimodal interaction that are reproducible and scalable.
•Create reward infrastructure including verifiable/programmatic rewards, reward model serving, LLM-as-judge pipelines, and defenses against reward hacking.
•Develop evaluation, monitoring, and debugging tooling to keep RL runs stable and diagnose convergence/throughput regressions and model behavior mid-run.

Pay and Benefits

Salary: USD 187,500 - 395,000 annually

Key Requirements

•Hands-on experience post-training LLMs with reinforcement learning methods (e.g., PPO/GRPO-family), including RLHF and RL from verifiable rewards.
•Extensive experience with distributed PyTorch training and parallelization for foundation models (e.g., FSDP, Tensor/Pipeline/Expert Parallel).
•Experience building RL environments, reward functions, verifiers, or evaluation harnesses for LLM agents, including sandboxed execution and multi-turn tool use.
•Strong familiarity with RL post-training frameworks and their system tradeoffs (e.g., veRL, OpenRLHF, TRL, Ray-based orchestration) plus rollout inference engines (vLLM, SGLang).
•Understanding of GPU clusters and communication libraries (NCCL, MPI) and how they behave under mixed training and inference workloads.

Experience:Multimodal AIFoundation modelsReinforcement learningLLM post-training

Skills:DebuggingCollaboration

Tech Stack:PyTorchFSDPNCCLMPIVLLMSGLangKubernetesRayTRLOpenRLHFVeRLPPOGRPORLHFRLVR

Company Brief

Luma

Develops AI-powered tools for capturing, editing, and rendering high-quality 3D scenes from photos and videos, enabling creators to generate photorealistic 3D assets and spatial experiences.

Industry: AR/VR & Spatial Computing

Website

Loading job details...

Preparing the role view and application actions.

Luma

3 days ago

Research Scientist / Engineer – Reinforcement Learning Infrastructure

✓ Verified Job

Canonical indexed version, validated from employer's careers page.

Source: Company careers pageValidated by: Fursa AI

Last checked: 2 hours agoStatus: Live

Job Summary

Location: Redwood City, London

Workplace: Hybrid

Employment Type: Full time

Job Function: Data Science & Machine Learning

Key Responsibilities

•Design, build, and scale distributed RL post-training systems for large multimodal models, orchestrating trainer, rollout, environment, and reward workloads across thousands of GPUs.
•Build and optimize high-throughput rollout generation by integrating inference engines (vLLM, SGLang) into the training loop and applying weight synchronization and asynchronous/off-policy schemes.
•Design and implement RL environments for agentic, multi-step tasks with sandboxed code execution, tool/computer use, and multimodal interaction that are reproducible and scalable.
•Create reward infrastructure including verifiable/programmatic rewards, reward model serving, LLM-as-judge pipelines, and defenses against reward hacking.
•Develop evaluation, monitoring, and debugging tooling to keep RL runs stable and diagnose convergence/throughput regressions and model behavior mid-run.

Pay and Benefits

Salary: USD 187,500 - 395,000 annually

Key Requirements

•Hands-on experience post-training LLMs with reinforcement learning methods (e.g., PPO/GRPO-family), including RLHF and RL from verifiable rewards.
•Extensive experience with distributed PyTorch training and parallelization for foundation models (e.g., FSDP, Tensor/Pipeline/Expert Parallel).
•Experience building RL environments, reward functions, verifiers, or evaluation harnesses for LLM agents, including sandboxed execution and multi-turn tool use.
•Strong familiarity with RL post-training frameworks and their system tradeoffs (e.g., veRL, OpenRLHF, TRL, Ray-based orchestration) plus rollout inference engines (vLLM, SGLang).
•Understanding of GPU clusters and communication libraries (NCCL, MPI) and how they behave under mixed training and inference workloads.

Experience:Multimodal AIFoundation modelsReinforcement learningLLM post-training

Skills:DebuggingCollaboration

Tech Stack:PyTorchFSDPNCCLMPIVLLMSGLangKubernetesRayTRLOpenRLHFVeRLPPOGRPORLHFRLVR

Company Brief

Luma

Develops AI-powered tools for capturing, editing, and rendering high-quality 3D scenes from photos and videos, enabling creators to generate photorealistic 3D assets and spatial experiences.

Industry: AR/VR & Spatial Computing

Website