Senior Performance Engineer | AI Infrastructure | Cambridge (Hybrid) |

Nobody quite knows where their compute budget is actually going until someone builds the model that tells them. That's this role. My client is a Cambridge-based non-profit that exists to stop different parts of the AI world quietly rebuilding the same infrastructure. Rather than a startup, a big enterprise, a government department and a university lab each working out GPU efficiency from scratch, they pool the hard problems and the expertise needed to solve them, so everyone moves faster. It's early days for the organisation but there's serious momentum and serious financial backing behind it already. They're hiring Performance Engineers at junior and senior level, to sit at the sharp end of that mission.

Day to day you would sit between the research and engineering teams, pulling real numbers off live training and inference runs rather than working from theory. From there, the job is building the models and calculators that turn those numbers into an actual answer: will this optimisation help, would a different accelerator be worth the spend, is this architecture change going to pay for itself. Those answers don't stay internal either; they shape what gets bought and how systems get built, for the organisation itself and for everyone else in the membership relying on that judgement.

What you'll bring:

• A degree in computer science, mathematics, or something adjacent
• A track record of building performance models or calculators (Python or spreadsheet-based) that actually forecast how a system will behave
• Hands-on GPU/accelerator code optimisation, CUDA or similar
• Genuine understanding of how LLMs and deep learning models run on real hardware, training versus inference, matrix multiplication, KV-caching, that level of detail
• Comfortable in profiling tools like Nsight or PyTorch Profiler, and monitoring stacks like Prometheus and Grafana
• Python for data work, Pandas and NumPy, plus general scripting

Nice to have rather than essential:

• A postgraduate degree and research background (publications welcome)
• Real depth on inference serving frameworks like vLLM
• A stats background
• Any open source or research contributions

Why look twice at this one:

It's a rare early seat at something with genuine backing and genuine ambition, where the work you do gets acted on rather than filed away. Competitive salary and pension, hybrid from a Cambridge office, and real exposure to people across the wider AI and academic scene.