Summary

Research Engineer responsible for building a reproducible benchmarking system for agentic and algorithmic LLM approaches. The role combines rigorous evaluation research with production engineering across sandboxed execution, distributed systems, and code-agent workloads, working in person from the London office.

Responsibilities

• Design unified agent evaluation systems
• Mine commits and traces for task suites
• Run sandboxed execution grading at scale
• Enforce contamination controls and stable baselines
• Compare agentic approaches using resolved-task quality
• Lead external benchmark co-publications
• Review quality claims before publication