AI Infrastructure Performance Engineer (Hybrid)

CommonAI Holdings Ltd; Cambridge; Hybrid; GBP 70,000 - 110,000; Full time

Benefits offered by this job

Pension

Benefits package

Cambridge office

Job summary

CommonAI CIC in Cambridge is seeking a researcher to design and run reproducible experiments on AI infrastructure, building measurement tools and observing real hardware and software behavior in LLM serving systems.

You will work with workload, simulation and uncertainty specialists to identify bottlenecks, explain performance variation and calibrate simulator models. Hybrid work with on-site Cambridge time; UK-wide applicants welcome.

Qualifications

• Experience designing reproducible experiments for AI hardware/software stacks.
• Ability to build and use measurement tools to collect traces, profiles and counters.
• Strong profiling skills across accelerator, memory and HPC workloads.

Responsibilities

• Design and run reproducible experiments from components to full LLM serving systems.
• Develop reusable measurement tools and collect traces, profiles and counters to characterise latency, throughput, memory behaviour and energy.
• Investigate bottlenecks and performance variation across layers, extend investigations through the stack.
• Ensure reproducible workload details and calibration data for simulator models.
• Collaborate with simulation and uncertainty specialists to choose measurements that resolve unknowns.

Skills

Experiment design

Measurement tools

Profiling expertise

AI hardware basics

Tools

Nsight

PyTorch Profiler

System profilers

Job description

CommonAI CIC in Cambridge is seeking a researcher to design and run reproducible experiments on AI infrastructure, building measurement tools and observing real hardware and software behavior in LLM serving systems.

You will work with workload, simulation and uncertainty specialists to identify bottlenecks, explain performance variation and calibrate simulator models. Hybrid work with on-site Cambridge time; UK-wide applicants welcome.