Software Engineer, ML Performance Optimization

About the role

Zoox seeks a Software Engineer to establish a dedicated discipline for ML Performance Optimization. This role defines a new function responsible for the end-to-end performance lifecycle of machine learning models deployed in Zoox robotaxis. You will own the workflow from initial intake through final deployment, ensuring that foundation models and inference pipelines execute with maximum speed and efficiency inside the vehicle compute stack. Success in this position requires close collaboration with Perception and Planner teams to redesign training procedures and streamline inference paths. The primary mission is to accelerate the deployment of autonomous driving capabilities by applying rigorous system-level tuning to every model. This position operates at the intersection of software engineering and applied research, building the infrastructure that allows other teams to ship advanced driving behaviors safely.

The Opportunity

You will lead initiatives to optimize ML performance across the entire Zoox stack. This involves working directly with current hardware and software techniques, including large-scale distributed training, quantization, model distillation, and pruning. You will partner with every autonomy team-Perception, Prediction, Planner, Simulation, and Collision Avoidance-to ensure that model performance aligns with real-world driving requirements. In this capacity, you will build and operate the foundational ML tools, development workflows, and serving systems that serve as force multipliers for internal customers. This role offers significant growth potential as Zoox expands robotaxi deployments and explores new ML domains. If you wish to deepen your knowledge of the ML infrastructure powering autonomous vehicles, additional details are available on our career site.

Key facts

-

Location: USA

-

Engagement: Full-time employment.

-

Compensation: Base salary of $183,226.00 for the 2024 year.

What you'll do

You will design standardized intake procedures for model metrics and hardware telemetry, capturing critical data before constructing analytical views. You will configure training infrastructure, specifically setting quantization parameters and distillation flows for our simulation clusters. You will establish formal review checkpoints to validate accuracy tradeoffs when pruning models intended for collision avoidance modules. You will coordinate model shipping activities, aligning conversion processes with vehicle integration schedules across multiple simulation environments. You will guide cross-functional partner interactions, sharing tool designs with Planner teams to improve route prediction accuracy. You will refine build processes, adjusting compiler options for Vision Language Models operating on edge compute devices. You will monitor review dashboards that track inference latency, proposing adjustments to vehicle logging pipelines as needed. You will drive ship documentation, recording configuration choices that affect distributed training experiments over time.

Requirements

You must possess hands-on experience with SOTA accelerators deployed in both data centers and vehicle computers. You require a deep understanding of distributed training patterns, quantization methods, and model distillation practices. You need familiarity with pruning techniques and their application to large Vision Language Models for efficiency gains. You should be proficient with compiler toolchains targeting GPU and specialized hardware for inference workloads. You must demonstrate the ability to collaborate effectively with Autonomy teams, including Perception, Prediction, Planner, and Simulation.

Nice to have

Prior experience presenting at conferences such as re:Invent, sharing insights on ML Infrastructure topics, is valued.

Skills & tools

SOTA accelerators, distributed training, quantization, distillation, pruning, VLMs, VLAs, simulation, collision avoidance, re:Invent.

Practical notes

Please About the company

Zoox is building the world's most advanced self-driving hardware and software solution. The efficiency demands of such a system require an expert fine tuning of both the compute hardware architecture as well as the algorithms and middleware that runs on it to achieve maximum throughput at the most optimal power levels.

About Zoox

Zoox is hiring for Software Engineer, ML Performance Optimization. The listing location is Foster City, CA.

This Software Engineer, ML Performance Optimization opening is posted for Foster City, CA.