Senior Staff Software Engineer, ML Performance, GPU
Sunnyvale, CA, USA · Senior Staff · Full-time · Posted Oct 19, 2026
- Experience
- 8 years in software development; 7 years leading technical project strategy, ML design, and end-to-end ML optimization. Preferred: 8 years with data structures and algorithms; 5 years in technical leadership; 3 years in a complex, matrixed organization.
- Education
- Bachelor’s degree or equivalent practical experience; preferred: Master’s degree or PhD in Engineering, Computer Science, or a related technical field.
About the job
Optimize GPU performance for Gemini and other LLM training and serving, from benchmarking and kernel development to scalable improvements across compilers and kernel libraries.
What you'll do
- Maintain LLM benchmarks, identify performance opportunities, improve XLA:GPU and Triton performance, and guide XLA releases.
- Work with DeepMind to onboard models and improve large-scale LLM training and serving on GPUs.
- Analyze GPU architectures, performance metrics, and fleet workloads to identify bottlenecks and implement efficiency improvements.
What they ask for
- Bachelor’s degree or equivalent practical experience.
- 8 years of experience in software development.
- 7 years of experience leading technical project strategy, ML design, and end-to-end Machine Learning optimization.
- Experience with modern GPU architectures and memory hierarchies; low-level GPU programming and performance engineering; and modern LLMs deployed on AI accelerators.
Nice to have
- Master’s degree or PhD in Engineering, Computer Science, or a related technical field.
- Experience with data structures and algorithms, technical leadership, complex matrixed organizations, or machine learning algorithms and tools.
Pay
$262000 - $364000 USD base salary
Tools
XLA:GPU, Triton, CUDA, CUTLASS, GPU
AI focus
GPU performance optimization for Gemini and other LLMs, including training and serving, benchmarking, model and data efficiency, quantization, compiler optimization, and custom kernels.
Model-written summary · the original posting is authoritative