Machine Learning Engineer, GPU Kernel and Runtime

The Waymo ML Infrastructure team accelerates Waymo’s mission, by building the best ecosystem for sustainably innovating and shipping ML powered intelligence. 

Research, Production, and the Hardware teams are our primary stakeholders and our work powers the development of the state of the art models in the areas of Perception and Trajectory planning that are core to our autonomous driving software. We enable our partners by offering the best in class solutions for the entire model development lifecycle. These solutions include understanding the model business goals and platform hardware characteristics, and codesign the models for the hardwares. These solutions are developed in close collaboration with teams at different modeling teams. Scale and efficiency are core tenets our infra follows.

We are looking for engineers with ML software & systems expertise to help build the next generation Waymo onboard ML inference engine for Waymo fundamental model. You'll work across the entire ML stack from the system perspective, from efficient deep learning models, model compression, ML software (e.g. JAX, XLA, Triton, and CUDA), to . You will be pleasantly challenged with deploying Waymo ML models on limited computation resources. In this hybrid role, you will report to the Senior Manager of Runtime and Optimization. 

You will:

• Collaborate with ML practitioners on models for perception, behavior prediction, and planning, to understand their models and accelerate them onboard through custom NVIDIA GPU kernel development.

• Deep dive into the NVIDIA ML software and runtime stack, from custom CUDA ops to the XLA:GPU compiler and low-level libraries. Analyze numeric behaviors, debug complex compiler issues, and ensure inference results are stable and consistent. Develop tools/system software for optimal resource usage, hardware efficiency, and platform reliability in an ML serving system.

• Analyze ML workload performance at the hardware level; apply manual and AI agent-assisted techniques and develop highly optimized, custom CUDA/Triton operator libraries tailored to Waymo's specific architectures.

• Build tools to benchmark, profile GPU execution, and productize deep learning models for a streamlined and robust onboard and offboard deployment.

You Have:

• B.S. or M.S. in CS, EE, Deep Learning or a related field

• 5+ years of industry experience on system performance, hardware-level GPU optimization, or ML compilers

• Strong C++ and CUDA programming skills

• Extensive experience in NVIDIA GPU Kernel development to accelerate deep learning models

• Proven debugging and optimization experience on the XLA:GPU compiler, as well as the NVIDIA runtime stack

• Passion for developing and optimizing ML software stacks for modern ML accelerator architectures (framework, runtime library, ML compiler, efficient deep learning etc.)

We Prefer:

• M.Sc or PhD in Computer Science, Mathematics or a related field.

• Strong Python programming skills.

• Experience with advanced NVIDIA profiling (e.g., Nsight Compute) and debugging (e.g. cuda-gdb) tools.

• Solid experience with designing, training and debugging deep learning models to achieve the highest scores/accuracies.

• In-depth knowledge of ML frameworks, ML compilers, and IRs (Triton, HLO, MLIR, CuTe DSL, cuTile) or modern ML system architectures.

• Role-Related Knowledge

• • C++ Coding

  • CUDA profiling & debugging

  • ML runtime optimization

  • Custom GPU kernel development