Distillation Lead

Waabi, founded by AI visionary Raquel Urtasun, is the leader in Physical AI. With a world-class team, we're unlocking the next era of autonomous transportation with technology that's powering commercial autonomous trucks and robotaxis. Waabi is backed by and partners with world leaders in AI, automotive, logistics, and deep tech.   With offices in Toronto, San Francisco, Dallas, and Pittsburgh, Waabi is growing quickly and looking for diverse, innovative and collaborative candidates who want to impact the world in a positive way. To learn more visit: www.waabi.ai Waabi’s Physical AI platform is powered by state of the art ML models which must be deployed efficiently across diverse use-cases, from onboard vehicle inference to large-scale simulation. As the Distillation Lead, you will own the strategy and execution for distillation across Waabi's AI stack, ensuring our most capable models run efficiently in every deployment context. You will partner closely with ML Platform, Infrastructure, Onboard Autonomy, and Simulation teams to deliver compressed models that meet the performance requirements of both real-time onboard systems and high-throughput simulation pipelines.   You will… Define and drive the technical strategy for model distillation and compression across Waabi's AI stack — spanning perception, world models, and planning — with an eye toward both onboard deployment and simulation use-cases. Design, implement, and scale state-of-the-art distillation and efficiency pipelines, which may include: Distillation for generative models (diffusion, autoregressive, flow-matching, video models) Quantization-aware training (QAT) and post-training quantization (PTQ) Knowledge distillation (feature-level, response-based, and relation-based) Structured and unstructured pruning and sparsification Low-rank factorization and efficient architecture design Speculative decoding and other inference-time efficiency techniques Collaborate closely with ML Platform, Infrastructure, Onboard, Autonomy, and Simulation teams to integrate compressed models into production pipelines and meet latency, memory, and throughput targets across deployment contexts. Define rigorous benchmarks and evaluation frameworks to characterize efficiency vs. quality trade-offs across models and hardware targets. Mentor and guide researchers and engineers working in the distillation and model efficiency space, setting a high technical bar and fostering a culture of rigorous experimentation. Champion best practices for model compression across the organization; disseminate knowledge through internal design reviews, documentation, and technical talks. Stay at the cutting edge of model efficiency research; contribute to the broader scientific community through publications and open-source contributions.   Qualifications: Deep distillation expertise: You have extensive hands-on experience designing and implementing distillation, quantization, pruning, and model compression techniques for large-scale neural networks, with demonstrated impact in production settings. Strong research and engineering foundation: A Bachelor's or Master's degree in Machine Learning, Computer Vision, Robotics, or a related field, or equivalent industry experience; relevant hands-on experience in model distillation and efficiency is what matters most. Expert Python and PyTorch (or JAX) skills with experience in large-scale distributed training. Technical leadership: You have a proven track record of setting technical direction and driving projects from conception to production. You inspire and elevate those around you through deep technical expertise and mentorship. Cross-functional collaboration: You have experience working closely with infrastructure, platform, and autonomy teams to deploy compressed models under real engineering constraints. Clear communicator: You can communicate complex technical trade-offs clearly to diverse audiences and drive alignment across research and engineering teams.   Bonus: Experience with hardware-aware optimization (TensorRT, ONNX, custom CUDA kernels, hardware-specific quantization). Publications at top-tier ML/CV venues (NeurIPS, ICML, CVPR, ICLR, ECCV) in model compression, efficient deep learning, or related areas. Experience distilling large generative models (diffusion models, LLMs, VLMs, or video models). Background in autonomous vehicles or robotics.