ML Systems Engineer, Inference Benchmarking

About CommonAICommonAI CIC is a non-profit organisation built around collaborative engineering and is backed by a mix of UK Government and private funding in order to design, build and deploy innovative AI systems. Our Scaling Inference Lab (SIL) is developing simulation grounded in real measurements to help the UK invest in promising AI infrastructure. We connect proposed hardware designs to realistic workloads, helping investors and engineers understand their potential, their limitations and the next experiment worth doing.

The Opportunity

Build the workloads and benchmarks that reveal how AI infrastructure performs on useful applications. You will develop reproducible LLM serving workloads, connecting application behaviour and requirements to physical measurements and SIL's simulation platform. The initial focus includes long-context inference and agentic workloads. You will investigate how request patterns, context lengths, caching, repeated model calls and interactions with tools shape the demands placed on a system.

Your benchmarks will make performance comparisons meaningful by specifying the work being done and the quality and service requirements it must meet. Working with measurement and simulation engineers and our partners, you will help preserve the workload behaviour that matters when moving between real systems and models. Your work will help engineers and investors understand which infrastructure choices suit particular applications, and how conclusions change when the workload changes.

What You Will Do

• Build and maintain reproducible LLM serving workloads, serving configurations and benchmark tools, including long-context and agentic workloads, using suitable existing software
• Characterise relevant behaviour such as request arrivals, prompt and output lengths, caching, batching, tool interactions and dependencies between requests
• Define performance measures alongside application-quality and service requirements, so comparisons reflect useful work under stated conditions
• Work with measurement engineers to reproduce workloads on physical systems, and with simulator engineers to develop workload representations that preserve the behaviour and dependencies that matter
• Investigate how changes in workloads and serving configurations affect conclusions, and keep the benchmark suite relevant as inference technology and workloads evolve
• Document datasets, model and software versions, configurations, provenance and permitted use so other engineers can reproduce and extend the work

Requirements

• Experience building or evaluating ML systems, inference services, performance benchmarks or comparable workload-driven systems
• Strong practical software skills for constructing reliable benchmark tools, analysing results and debugging interactions between applications and infrastructure
• Understanding of LLM inference and an interest in how serving choices affect performance and output quality
• Care in designing reproducible comparisons and interpreting results against stated application requirements
• Practical experience using AI coding agents, with the judgement to validate generated code, data transformations and evaluations

Experience with vLLM, SGLang or similar serving software, long-context workloads, agentic applications, trace analysis or quality evaluation would be valuable. Depth in a subset is sufficient. The initial remit is LLM serving; other forms of inference are possible later extensions.

Why join us

• Help build UK AI capability and improve how investment and compute serve the public good
• Build a network spanning UK hardware SMEs, major technology companies, venture and public investors, and leading universities
• Shape a growing team with startup pace and close academic links
• Use frontier…