Principal Software Engineer, ML Flywheel Technical Lead

Software Engineering builds the brains of Waymo's fully autonomous driving technology. Our software allows the Waymo Driver to perceive the world around it, make the right decision for every situation, and deliver people safely to their destinations. We think deeply and solve complex technical challenges in areas like robotics, perception, decision-making and deep learning, while collaborating with hardware and systems engineers. If you're a software engineer or researcher who's curious and passionate about Level 4 autonomous driving, we'd like to meet you.

Every mile driven by a Waymo car is a unique and valuable piece of data that can be leveraged into improving our machine learning models, and ultimately the safety and capabilities of the Waymo Driver. Charting the path from collection of a piece of data, to curation, (auto-)labeling, model training and evaluation, all the way to model deployment and monitoring is a process of continuous scaling and quality improvement of the entire ML lifecycle. The Area Technical Lead for the Waymo machine learning flywheel is responsible for architecting, connecting, automating and improving the entire span of this self-improvement process in close collaboration with Waymo's infrastructure, modeling and evaluation teams. They are directly accountable to Waymo's leadership for the organization's ML data strategy and its impact on Driver quality.

You will report directly to our Distinguished Engineer, Foundation Models.

You Will:

• Architect a path towards every autonomous mile driven by a Waymo car to be automatically incorporated into an automated data-driven self-improvement loop for the Waymo Driver.
• Enable a data flywheel to serve the demands of scalable pre-training, post-training targeted to relevant critical behaviors, as well as Driver simulation and validation.
• Enable a modeling flywheel to efficiently consume that data to reliably generate updated models that are validated and deployable with minimal human toil.
• Coordinate cross-functional efforts in partnership with data and ML infrastructure teams, resource planning, logging infrastructure, modeling and validation teams to accelerate the velocity, impact and leverage of driving data on the Waymo Driver.
• Act as the steward of data quality, by providing tooling and metrics to evaluate the impact of mining, selection and curation on the modeling pipeline performance.
• Articulate the strategy for incorporating diverse data sources, including third-party and synthetic data into that improvement flywheel.
• Drive innovation across all axes of performance and efficiency, from Driver quality, to scalability, cost, engineering velocity, model architecture and performance.

You Have:

• Master's degree or PhD in Computer Science, Engineering, or a related technical field
• 10+ years of experience in ML model development, and you have 2+ years experience with large-scale vision, video, or multi-modal foundation model development and their integration in end-to-end models
• 6+ years of experience in ML-driven production systems that develops models with large-scale data, training, evaluation, and deployment
• 6+ years of experience in a technical leadership role leading technical teams and setting technical directions in large ML Engineering organizations
• Demonstrated expertise in large-scale machine learning and its key components: pre-training, post-training, and validation. Deep understanding of both infrastructure and quality aspects.
• Track record of architecting and standing up a machine learning flywheel at scale for mission critical applications.
• Experience with driving model quality improvements through systematic data scaling and curation.
• Communication and interpersonal skills. Ability to inspire, influence and coordinate across functions and disciplines.

We Prefer:

• PhD in computer science.
• Experience in multi-modal LLM model development, and their infrastructure.
• Familiarity with multi-task, end-to-end models and their development challenges.