Staff ML Performance Engineer, Performance Architecture

Overview

In this role you will optimize ML workloads across training and cloud inference, driving efficiency gains and scalable performance. You’ll collaborate with Research, model teams and platform groups to embed performance engineering into development cycles. Your work will focus on profiling, bottleneck removal, and reusable cross-target optimisations that boost MFU, throughput and utilization. You’ll help shape cloud GPU strategy and establish benchmarking that tracks progress. This role offers an opportunity to impact autonomous driving at scale with a performance-first mindset.

Pay / Benefits

• Meaningful equity
• Relocation support and visa sponsorship
• Hybrid working with core hours
• Learning and development budgets
• Comprehensive benefits (health insurance, dental, parental leave)
• Wellbeing programs and team socials

Responsibilities

• Profile ML workloads to identify bottlenecks in training and cloud inference
• Design and implement cross-target optimisations (kernels, data loaders, frameworks like Triton)
• Build reusable benchmarking tools and track efficiency gains
• Quantify improvements in MFU, throughput, and utilization
• Collaborate with Research and model teams to embed performance into development cycles
• Partner with platform teams to inform cloud GPU hardware strategy
• Foster a culture of performance optimization across teams

Key requirements

• 10+ years of experience in performance engineering across ML systems, GPU compute infrastructure or distributed platforms
• Experience optimizing large-scale GPU workloads for training and/or inference
• Proven ability to write, report and track performance benchmarks
• High-quality, well-structured, tested Python code
• Bachelor in Computer Science (or equivalent)

• collaborative communication
• problem-solving mindset
• data-driven decision making
• ML systems performance engineering
• GPU compute infrastructure and distributed platforms
• profiling tools (e.g., system profilers)

…