Research Engineer – Performance Reinforcement Learning

About the role

Anthropic is looking for a Research Engineer to join its Code RL team. You will focus on improving the ability of large language models to write correct, high‑performance code for accelerators, turning hardware performance insights into learnable tasks and signals.

Key responsibilities

• Invent, design, and implement reinforcement‑learning environments and evaluation pipelines.
• Run experiments, analyze results, and shape the research roadmap.
• Integrate successful prototypes into large‑scale training runs.
• Collaborate with researchers, engineers, and performance‑engineering specialists across Anthropic.

Required profile

• Strong expertise with accelerator technologies such as CUDA, ROCm, Triton, or Pallas.
• Proficiency in ML framework programming using JAX or PyTorch.
• Experience working across the stack – from kernels to model code to distributed systems.
• Ability to balance exploratory research with production‑grade engineering.
• Passion for safe, beneficial AI development.

Required skills

• CUDA
• ROCm
• Triton
• Pallas
• JAX
• PyTorch
• Reinforcement learning
• Large language model training methodologies
• Distributed systems

What we offer

• Opportunity to influence cutting‑edge AI models used in Anthropic’s Claude series.
• Collaboration with top‑tier alignment, red‑team, and production groups.
• Access to state‑of‑the‑art hardware and research infrastructure.