Overview
As a Staff ML Performance Engineer, you will optimize ML inference for edge accelerators and GPUs to enable Wayve’s first driving product. You will help shape the technical direction for production-ready models running on in-vehicle compute, across ML systems, compilers, runtimes, and embedded deployment. Expect to tackle bottlenecks in the full inference stack and drive measurable performance gains. You’ll work closely with model developers and cross-functional teams to raise performance standards and deliver scalable, edge-ready solutions. This role offers the chance to influence the hands-on deployment of transformer-based models in a safety-critical domain and contribute to Wayve’s core,
Pay / Benefits
- hybrid working policy in London
- office-based collaboration
- inclusion-focused interview process
- global impact on autonomous driving
- opportunity to shape core technology
- career development support
Responsibilities
- Profile and pinpoint bottlenecks across the full inference stack (model graph, compiler/runtime, kernel execution, memory movement) and deliver measurable improvements
- Implement and validate optimisations in compilers, runtimes, and/or kernels (e.g. operator fusion, scheduling, quantisation-aware performance, custom kernels)
- Build robust benchmarking and regression testing to ensure performance improvements hold across models, devices, and software releases
- Optimise for multiple targets (e.g. NVIDIA Orin/Thor, Qualcomm) and work with teams to support these in a maintainable way
- Collaborate with model developers to influence architecture and training/deployment decisions that affect on-device performance
- Contribute to technical roadmaps and tooling and help raise the standard of performance engineering across the team
Key requirements
- Proven experience improving performance in production systems with tight constraints (latency, memory, bandwidth, power/thermal, or cost)
- Strong proficiency with at least one relevant stack/toolchain (e.g. TensorRT, CUDA, Qualcomm QNN, Triton, OpenCL) and confidence learning adjacent frameworks quickly
- Comfort operating at multiple levels of abstraction — from high-level model behaviour down to low-level kernel/runtime execution
- Strong software engineering fundamentals (debugging, profiling, testing, and maintainable code)
- Clear communicator and collaborative teammate; able to align multiple stakeholders on performance trade-offs and priorities
- Clear communicator
- Collaborative teammate
- Able to align stakeholders on trade-offs
- TensorRT
- CUDA
- Qualcomm QNN
…