Senior NPU Kernel Operator Engineer

Senior AI Accelerator Performance Engineer (Kernels & Operators)

Location: Flexible, to be discussed

Salary: Highly competitive, based on experience

Sector: AI Hardware / High-Performance Computing

The Opportunity

I'm partnering with a forward-thinking technology business building next-generation AI compute. They're looking for a Senior Performance Engineer to design, write and tune the low-level kernels and operators that sit at the heart of their AI stack.

You'll be the bridge between algorithms and silicon: taking the mathematical building blocks of modern AI models and turning them into implementations that squeeze every bit of performance out of the hardware. Expect to work closely with architecture, compiler and software teams, with real influence over how workloads run on the platform.

This suits a seasoned engineer who enjoys working close to the metal and gets genuine satisfaction from making fast code faster.

What You'll Bring

Must-haves

• Degree (BSc, MSc or PhD) in Computer Science, Computer/Electrical Engineering or a similar field
• 5+ years in a performance-focused role, such as HPC, accelerator or GPU programming, compilers, embedded systems or processor optimisation
• Excellent C and C++
• Deep understanding of computer architecture: memory hierarchy, data movement, parallelism and throughput
• A track record of speeding up performance-critical maths or compute workloads
• Comfort profiling code, pinpointing bottlenecks and fixing them methodically
• A problem-solver who's happy working across the hardware/software boundary

Strongly preferred

• Hands-on experience with low-level parallel or accelerated programming models
• Practical use of tiling, memory management, vectorisation, instruction-level tuning and concurrency
• Experience optimising numerical or ML workloads specifically
• Ability to reason about latency, bandwidth, utilisation and efficiency trade-offs
• Experience debugging tricky correctness, precision and performance issues

Nice to have

• CUDA, OpenCL or comparable parallel programming frameworks
• Exposure to compilers, code generation or kernel autotuning tools
• Familiarity with common ML, linear algebra or image processing operators
• Knowledge of low-precision and quantised number formats
• Experience with custom silicon, FPGAs, simulators, emulators or bring-up on real hardware
• Cross-team collaboration with architecture and infrastructure groups
• Mentoring others or setting up performance engineering best practice

Interested or know someone who might be? Get in touch for a confidential conversation.

Darwin Recruitment is acting as an Employment Agency in relation to this vacancy.