ML Infrastructure Engineer, Distributed Systems

Speedrun Talent Network; Greater London; Hybrid; GBP 110,000 - 170,000; Full time

Job summary

Harvey is hiring a Member of Technical Staff, Infrastructure to scale the post-training stack for legal-focused models. You’ll own distributed training, RL rollout systems, and environments where models learn to interact with tools and documents.

You’ll work with a small high-bar team of engineers and researchers to make training faster and more reliable. You’ll partner with researchers to turn experimental workflows into dependable systems while keeping room for rapid experimentation, with the

Qualifications

• Experience building ML infrastructure and distributed systems.
• Strong Python and backend engineering skills with ML framework familiarity.
• Experience with distributed training or inference and bottleneck diagnosis.
• Understanding post-training workflows; RL rollouts or tool execution is a plus.
• Practical judgment to ship quickly while aligning with research.

Responsibilities

• Scale post-training stack for SFT and RL on open-weight models.
• Build distributed rollout infrastructure coordinating inference and tool use.
• Create reproducible agent environments for training and evals.
• Own performance, reliability, and observability of training runs.
• Identify bottlenecks to shorten research-to-result cycles.
• Build internal tools for data management, evals, and experiments.
• Integrate models into production with latency and cost awareness.
• Collaborate with researchers to stabilize workflows and enable experimentation.

Skills

ML infrastructure

Distributed systems

Python engineering

Research tooling

Production readiness

Tools

PyTorch

JAX

Job description

Harvey is hiring a Member of Technical Staff, Infrastructure to scale the post-training stack for legal-focused models. You’ll own distributed training, RL rollout systems, and environments where models learn to interact with tools and documents.

You’ll work with a small high-bar team of engineers and researchers to make training faster and more reliable. You’ll partner with researchers to turn experimental workflows into dependable systems while keeping room for rapid experimentation, with the