Machine Learning Engineer, Performance Tooling

The role

You’ll join the AI Performance Tooling team within Wayve’s AI Performance organization, which makes model training and inference faster, more efficient, and more predictable across cloud and embedded hardware. Our mission is to enable data-driven AI performance decisions across priority workloads and hardware targets: Measure performance teams can trust; Monitor trends and catch regressions; Predict the cost of changes before we run them; Advise on bottlenecks and prioritized opportunities. You’ll build tools that reason across the AI stack — models and operators, compilers, runtimes, accelerators, and distributed training infrastructure — turning profiling data into a clear picture of where time, memory, power, and compute go, and what proposed changes will do to latency, throughput, compute spend, and capacity. You’ll work closely with model, compiler, runtime, platform, and hardware teams, bringing a cross-stack view that turns measurement into clear recommendations.

Key responsibilities

• Design and build reliable, self-service performance tools that scale across models, hardware targets, and development workflows.
• Shape how Wayve measures and predicts AI performance, and set the standards other teams build on.
• Model theoretical peak for a platform, compare with achieved performance, and pinpoint where efficiency is lost at layer and op level.
• Predict latency, memory, utilization, and compute cost of a model or recipe change before spending compute.
• Own monitoring and regression alerting across model builds and training runs.
• Work with training and runtime engineers to set performance targets and make the case with data.

About you

In order to set you up for success as a Software Engineer, AI Performance Tooling at Wayve, we’re looking for the following skills and experience.

Essential

• Deep, hands-on performance engineering in complex systems: profiling, roofline analysis, latency and throughput optimization, and root-causing what limits a workload.
• A track record of owning a tool or service end to end — design, delivery, and adoption by other teams.
• Strong Python skills, and comfort profiling and instrumenting large production codebases.
• Hands-on experience developing deep learning models with PyTorch.
• Data analysis skills to turn noisy measurements into conclusions you can defend.
• Judgment to turn an ambiguous performance question into a measurable one, and to prioritize what matters.
• Quantitative communication clear enough to influence another team’s priorities.