Research Engineer, Waabi

• Strong software engineering and implementation: You have very strong Python & PyTorch (or JAX) skills; strong software-engineering fundamentals, and extensive experience with distributed training and large-scale model deployment
• Demonstrated technical impact: You have a Master's degree in Computer Vision, Machine Learning, Robotics, or a related field, or equivalent industry experience in model development and scaling
• Expert domain knowledge: You have built and deployed generative or predictive models of the physical world, focusing on scale, efficiency, and robustness for real-world applications
• Team player: You have worked in a close-knit team of researchers and engineers and have strong communication to deliver successful projects in a fast-paced environment

Desirable

• Waabi’s autonomous-driving stack is powered by Waabi World, which delivers realistic, scalable, controllable, and efficient simulation
• As a Research Engineer in the World Models team, you will develop algorithms and productionize the next generation of World Models that can reason about complex, dynamic 4D environments
• This role focuses on large-scale world models for temporal reasoning and generation, including video models, multimodal generative models, LLM/VLM/VLA models, and predictive models of traffic participants and scenes
• Your work will directly power Waabi World’s ability to model future evolution, synthesize realistic safety-critical scenarios, and provide rich generative priors for downstream planning, testing, and training
• Design, implement, and scale state-of-the-art generative and predictive world-modeling systems:
• Video generation and prediction
• Latent diffusion / autoregressive / flow-matching models
• Multimodal foundation models for driving scenes
• LLM / VLM / VLA methods for scene understanding, reasoning, and control
• Generative scenario modeling and controllable simulation
• Model distillation
• Collaborate closely with Research Scientists to translate cutting-edge model prototypes into robust, large-scale, distributed training and inference pipelines
• Optimize model training and inference for efficiency, speed, and reliability on large-scale datasets
• Build large scale data pipelines to build high quality datasets for training
• Ensure the quality, stability, and maintainability of the world model codebase and infrastructure
• Stay on top of emerging advances in generative AI, distributed systems, and efficient model deployment in robotics