Research Scientist / Engineer – Reinforcement Learning Infrastructure
Luma · Remote · Remote, EU ·
COMPENSATION
€•• – ••k
Listed on the employer's job board (Ashby)
- Fully remote role
- Distributed RL systems
- Thousands of GPUs
- LLM post-training
- Senior level
You will design and scale the systems that enable reinforcement learning for frontier models. This includes coupling policy optimization with large fleets of inference workers and building the reward and verification systems that convert model behavior into learning signals. The work requires solving full-loop systems problems where training, rollout generation, and environment execution run concurrently across massive GPU clusters.
What you'll do
- Design and scale distributed RL post-training systems across thousands of GPUs
- Build high-throughput rollout generation with asynchronous and off-policy schemes
- Create scalable RL environments for agentic and multi-step tasks
- Develop reward infrastructure including verifiable rewards and LLM-as-judge pipelines
- Build tooling for evaluation, monitoring, and debugging of large RL runs
- Collaborate with researchers to turn new post-training ideas into production runs
What you'll need
- Hands-on experience post-training LLMs with RL at meaningful scale
- Extensive distributed PyTorch training and parallelism experience
- Experience building RL environments, reward functions, or evaluation harnesses
- Deep familiarity with RL post-training frameworks like veRL or OpenRLHF
- Strong understanding of GPU clusters, networking, and communication libraries
Nice to have
- Experience running RL training across 100+ GPUs
- Knowledge of containerization and orchestration with Kubernetes or Ray
- Research or open-source contributions in RL for LLMs
Stack
- PyTorch
- vLLM
- SGLang
- veRL
- OpenRLHF
- TRL
- Ray
- NCCL
- MPI
- Kubernetes
How you'll work
- Fully remote position
- Work with distributed GPU clusters