Machine Learning Systems Engineer, Reinforcement Learning Engineering

Conversant - San Francisco, CA

Full-time

San Francisco, CA

Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services

About the position

As an ML Systems Engineer on the RL Engineering team at Anthropic, you will play a pivotal role in developing and enhancing the systems that train AI models like Claude. Your primary focus will be on building, maintaining, and improving the algorithms and infrastructure that support our researchers in their mission to create advanced AI capabilities. This position requires a strong emphasis on performance, robustness, and usability to ensure rapid progress in AI research and development.

Responsibilities

• Build, maintain, and improve algorithms and systems for training AI models.
• Enhance the performance, reliability, and usability of training systems.
• Support and empower the research team in their AI development efforts.
• Profile the reinforcement learning pipeline to identify improvement opportunities.
• Develop a system for launching training jobs in a test environment to quickly detect issues.
• Modify finetuning systems to accommodate new model architectures.
• Create instrumentation to address Python GIL contention in training code.
• Diagnose and resolve issues causing slowdowns in training runs.
• Implement stable and efficient versions of new training algorithms.

Requirements

• 2+ years of software engineering experience.
• Experience working on systems and tools that enhance productivity.
• Results-oriented with a flexible and impactful approach.
• Willingness to take on tasks outside of the job description.
• Enjoyment of pair programming and collaborative work.
• Interest in learning more about machine learning research.
• Awareness of the societal impacts of AI work.

Nice-to-haves

• Experience with high performance, large scale distributed systems.
• Familiarity with Kubernetes.
• Proficiency in Python programming.
• Background in machine learning.
• Experience implementing LLM finetuning algorithms, such as RLHF.