Edge ML Inference Performance Lead

Wayve; London (KY); On-site; USD 149,000 - 203,000; Full time

Job summary

Wayve in London is seeking a Staff ML Performance Engineer to optimise inference for edge accelerators and GPUs, enabling our first driving product. You will help define the technical direction to run large transformer models reliably on in-vehicle compute, spanning ML systems, compilers, runtimes, kernels and embedded deployment.

You will collaborate with model developers to influence architecture and deployment choices, build benchmarking and tooling, and deliver measurable performance

Qualifications

• Production ML performance experience under tight constraints (latency, power, memory).
• Familiar with modern ML inference stacks and edge accelerators.
• Strong debugging, profiling, and maintainable code practices.

Responsibilities

• Identify, implement and validate ML compiler, runtime and kernel optimisations (fusion, scheduling, quantisation, custom kernels).
• Profile bottlenecks across the full inference stack and deliver measurable improvements.
• Build benchmarking and regression tests to ensure performance across models and devices.
• Develop and optimise for multiple platforms (e.g., NVIDIA Orin/Thor, Qualcomm) with cross-functional teams.
• Collaborate with model developers to influence architecture and deployment decisions for on‑device performance.
• Contribute to roadmaps and tooling to raise performance engineering standards.

Skills

TensorRT

CUDA

Triton

ONNX

Python

C++

Profiling

Performance optimization

Tools

NVIDIA Orin/Thor

Qualcomm QNN

OpenCL

MLIR

CUDA Toolkit

Job description

Wayve in London is seeking a Staff ML Performance Engineer to optimise inference for edge accelerators and GPUs, enabling our first driving product. You will help define the technical direction to run large transformer models reliably on in-vehicle compute, spanning ML systems, compilers, runtimes, kernels and embedded deployment.

You will collaborate with model developers to influence architecture and deployment choices, build benchmarking and tooling, and deliver measurable performance