About the job
In Semantics, our team's mission is to create the highest-fidelity, most comprehensive offboard perception autolabels at a massive scale, serving as the foundation for training and validating the AV stack. We are an advanced ML and engineering team that leverages state-of-the-art computer vision, deep learning, and generative AI to automatically analyze driving logs, generate rich scene understanding, and power the data engine that enables Waymo to scale safely and efficiently.
Responsibilities
Develop and train state-of-the-art computer vision / multimodal models (e.g., Gemini) to extract the rich semantic information (e.g., object attributes, scene properties, interaction dynamics) required by the AI agent.
Design and implement a scalable AI agent framework that integrates large foundation models (e.g., Gemini) with the outputs of our perception models and internal knowledge bases.
Develop and apply Fine-tuning and Reinforcement Learning (RL) techniques to create a "data flywheel," continuously improving the system's captioning and reasoning abilities through automated feedback.
Develop and prototype novel prompting strategies for Vision-Language Models (VLMs) to elicit complex, causal reasoning about driving scenarios.
Collaborate closely with the ML Infra, Perception, Behavior, and AI Foundation teams to define data requirements and integrate the captioning system into the broader ML development lifecycle.
Own the full system lifecycle, from advanced model development and prototyping to production deployment and scaling for massive data generation
Qualifications
Minimum
Master’s degree in Computer Science, or a related technical field.
4+ years of hands-on experience training and shipping deep learning models for computer vision tasks (e.g., detection, segmentation, video understanding) using Python and frameworks like PyTorch, JAX, or TensorFlow.
1+ years of demonstrated experience working with large language models (LLMs) or vision-language models (VLMs) in areas such as fine-tuning, prompting, or Retrieval-Augmented Generation (RAG).
Strong software engineering fundamentals, including designing scalable and reliable systems.
Experience building and managing large-scale data processing pipelines for ML training.
Proven ability to work autonomously and lead complex technical projects in a fast-paced R&D environment.
Preferred
PhD in Computer Science, or a related technical field.
Publication record in top-tier AI conferences (e.g., NeurIPS, ICML, ICLR, CVPR).
Hands-on experience with Reinforcement Learning, especially RLHF, RLAIF, or applying RL to language/agentic tasks.
Experience with modern techniques in self-supervised, weakly-supervised, or multi-task learning for perception.
Experience building with AI agent frameworks (e.g., LangChain, LlamaIndex) or developing autonomous agentic systems.
Familiarity with the challenges of multimodal perception in robotics or autonomous driving.
A track record of impactful cross-functional collaboration.