Senior/Staff Software Engineer, ML Training Performance

Zoox is on a mission to reimagine transportation and ground-up build autonomous robotaxis that are safe, reliable, clean, and enjoyable for everyone. We are still in the early stages of deploying our robotaxis on public roads, and it is a great time to join Zoox and have a significant impact in executing this mission. The ML Platform team at Zoox plays a crucial role in enabling innovations in large-scale Foundation models, VLMs, and VLAs to make autonomous driving as seamless as possible.

The Opportunity Are you excited to drive our ML Performance Optimization initiatives and make our ML models that enable autonomous driving as fast and efficient as possible? You will get to work with SOTA accelerators, cutting-edge techniques in distributed training, quantization, distillation, and pruning, among other things, working closely with all the Autonomy teams within Zoox - Perception, Prediction, Planner, Simulation, Collision Avoidance, and have the opportunity to significantly push the boundaries of how ML is practiced within Zoox.

We build and operate the base layer of ML tools, model development, and serving systems that our applied research teams use for in- and off-vehicle ML use cases. You will work alongside a team of strong software engineers and act as a force multiplier for our internal customers. This team has many growth opportunities as we expand our robotaxi deployments and venture into new ML domains. If you want to learn more about our stack behind autonomous driving, please look here. If you want to learn more about our ML Infrastructure, here is one of our past talks at re:Invent.

Lead the design, implementation, and operation of training performance optimizations (data/tensor/pipeline parallelism, mixed precision, gradient checkpointing, GPU kernels) to scale VLM, VLA, and Foundation model training. Own the technical strategy and multi-quarter roadmap for RL framework and RL performance optimization for driving foundation models. Drive profiling and root-cause analysis of throughput bottlenecks across the compute, compiler, framework, and cluster-networking stack. Partner with ML researchers, data engineers, and hardware engineers to define requirements and make architectural decisions for large-scale training infrastructure. Build custom kernels and benchmarks that track and validate performance gains across the stack. Mentor engineers and raise the team’s technical bar through design reviews and best practices.

Note: You do not have to meet all the requirements below to be considered for this position: 7+ years of total experience , including 4+ years on large-scale model training platforms, with a record of leading projects end to end. Deep experience with distributed training frameworks (Megatron-LM, TorchTitan) and efficiently utilizing GPU/accelerator clusters at scale. Experience with RL frameworks (Verl/MILES) and algorithms (GRPO, PPO). Experience with collective communication libraries (e.g., NCCL) and optimizing communication/compute overlap in multi-node training. Strong command of profiling tools (NVIDIA Nsight, PyTorch Profiler, or XLA/JAX profiling) to find and resolve training bottlenecks. Proficiency in Python or C++. Experience with ML compilers, performance tooling, or hardware/software co-design (GPU/TPU/accelerator architectures). Ability to influence across teams and communicate technical tradeoffs to both researchers and engineers.