Latest AI JobsBrowse Jobs
Luma AI logo

Research Scientist / Engineer – Reinforcement Learning Infrastructure

Luma AIRemote, EUResearch & AI
Apply on company site
Salary
Not listed
Job type
Full-time
Work mode
Remote
Experience
Senior
Posted
8 Oct 2026
Status
Open · checked 7d ago

About this role

Remote within the EU, Luma AI seeks a Research Scientist or Engineer for Reinforcement Learning Infrastructure to construct distributed reinforcement learning systems operating at frontier scale. This position suits an individual with deep practical expertise in managing large GPU clusters, post-training large language models, and scaling asynchronous rollout pipelines.

Role Overview

The Research Scientist or Engineer focuses on building the infrastructure that integrates policy optimization, inference workers, agentic environments, and reward systems. The candidate addresses full-loop systems challenges encompassing training, rollout generation, environment execution, and reward computation across thousands of GPUs.

Key Responsibilities

Design, construct, and scale distributed reinforcement learning post-training architectures that coordinate trainers, rollouts, environments, and reward workloads across thousands of GPUs.
Develop high-throughput rollout generation systems by integrating inference engines such as vLLM and SGLang, weight synchronization, and asynchronous or off-policy schemes.
Create reproducible reinforcement learning environments for agentic, multi-step tasks including sandboxed code execution, tool use, computer use, and multimodal interaction scaling to millions of episodes.
Construct reward infrastructure that incorporates verifiable rewards, reward-model serving, LLM-as-judge pipelines, and defenses against reward hacking.
Implement evaluation, monitoring, and debugging tools to maintain stability during large-scale training runs.
Collaborate with researchers to improve training efficiency and stability while moving new post-training concepts into production.

Requirements

Practical background in post-training large language models using reinforcement learning methods like PPO, GRPO, RLHF, and RLVR at significant scale.
Comprehensive knowledge of distributed PyTorch training and parallelism techniques including FSDP, Tensor Parallelism, Pipeline Parallelism, and Expert Parallelism for foundation models.
Experience designing reinforcement learning environments, reward functions, verifiers, or evaluation harnesses tailored for language model agents.
Familiarity with reinforcement learning post-training frameworks such as veRL, OpenRLHF, TRL, and Ray orchestration, alongside inference engines like vLLM and SGLang.
Understanding of GPU clusters, networking, and communication libraries like NCCL and MPI under mixed training and inference workloads.

Nice to Have

Background running reinforcement learning training across more than 100 GPUs using asynchronous or disaggregated architectures.
Proficiency in containerization and orchestration platforms like Kubernetes and Ray for managing large environment fleets and sandboxed workloads.
Documented research contributions in reinforcement learning for language models or open-source involvement with reinforcement learning training frameworks.

Skills and Tools

PyTorch
vLLM
SGLang
veRL
OpenRLHF
TRL
Ray
NCCL
MPI
Kubernetes

About the Company

Luma AI focuses on creating unified general intelligence capable of generating, understanding, and operating within the physical world. The organization emphasizes multimodality and believes vision represents the critical next step beyond traditional language models. Luma AI operates as an equal opportunity employer.

How to Apply

Interested candidates should apply directly on the official company careers page.

Skills and tech stack

PyTorchvLLMSGLangveRLOpenRLHFTRLRayNCCLMPIKubernetesLLMLLMsReinforcement Learning

Questions candidates ask

Where is the position located?

The role is remote within the EU.

What core technical frameworks are required?

Candidates must have experience with distributed PyTorch training, reinforcement learning post-training frameworks like veRL, OpenRLHF, TRL, and Ray, as well as inference engines such as vLLM and SGLang.

What scale of hardware infrastructure will the engineer work with?

The engineer will design and scale systems that orchestrate workloads across thousands of GPUs.

What is the primary mission of Luma AI?

Luma AI aims to build unified general intelligence that can generate, understand, and operate in the physical world, with a strong focus on multimodality.

How should candidates apply for the job?

Candidates should submit their application through the official Luma AI careers page.

Ready to apply?

The application is on the official Luma AI careers page.

Apply on company site