RL Research Engineer - Hybrid, Safe AI Systems

Anthropic Limited; New York (NY); Hybrid; USD 500,000 - 850,000; Full time

Job summary

Anthropic is seeking a Research Engineer in Reinforcement Learning to advance the capabilities and safety of large language models. You will blend research and engineering to implement novel approaches and influence research direction, focusing on agentic models via tool use and improving reasoning with math and prototyping internal productivity tools.

Join a collaborative team of researchers and engineers to develop scalable RL infrastructure, training environments, evaluation methods, and

Qualifications

• Proven experience in ML research or equivalent industry work.
• Strong Python programming and async/concurrent systems experience.
• Experience with ML frameworks (PyTorch, TensorFlow, JAX).
• Ability to balance research exploration with engineering implementation.
• Experience with pair programming and collaboration.
• Commitment to building safe and beneficial AI systems.

Responsibilities

• Architect and optimize reinforcement learning infrastructure and distributed experiment management.
• Design, implement, and test training environments and evaluation methodologies for RL agents.
• Collaborate across research and engineering to accelerate AI research and deployment.
• Develop automated testing frameworks and scalable APIs for RL workflows.

Skills

Python

Async/Concurrent

Trio

PyTorch

TensorFlow

JAX

ML Research

Pair programming

Code quality

Systems design

Communication

AI safety

Education

Bachelor's degree

Tools

Kubernetes

Rust

C++

Job description

Anthropic is seeking a Research Engineer in Reinforcement Learning to advance the capabilities and safety of large language models. You will blend research and engineering to implement novel approaches and influence research direction, focusing on agentic models via tool use and improving reasoning with math and prototyping internal productivity tools.

Join a collaborative team of researchers and engineers to develop scalable RL infrastructure, training environments, evaluation methods, and