Research Engineer, Infrastructure

Role overview

Build the systems that support post-training and evaluation of open-weight models for complex professional workflows. Working closely with researchers, you will own infrastructure spanning distributed training and rollouts, reproducible agent environments, internal research tools, and the path from experimental models to production.

Responsibilities - Scale supervised fine-tuning and reinforcement learning workflows for leading open-weight models. - Build distributed rollout systems that coordinate inference, tool execution, and collection of training experience for long-horizon tasks. - Create reproducible environments where models can use documents, tools, and external systems during training and evaluation. - Improve training-run performance and reliability through GPU utilization, checkpointing, failure recovery, and observability. - Develop data management, evaluation, and experiment tools for preparing datasets, reviewing results, and launching and tracking jobs. - Find bottlenecks across training and inference, and integrate research models into production with attention to serving, latency, cost, and online evaluation. - Partner with researchers to make experimental workflows dependable while preserving room to iterate.

Requirements - Experience building and operating ML infrastructure, distributed systems, or research tooling, with ownership of reliability and performance. - Strong Python and general software engineering skills across backend services, data systems, and internal tools. - Familiarity with PyTorch, JAX, or a similar machine learning framework. - Experience with distributed training or inference and diagnosing problems across the stack. - Practical judgment about building versus adopting tools, and ability to work closely with researchers as requirements change.

Nice to have - Understanding of post-training workflows, especially reinforcement learning rollouts, agent environments, or sandboxed tool execution.

Benefits and work setup - Hiring is open across multiple experience levels.