Zoox · posted · checked
Machine Learning Engineer – Vision Foundation Models
PythonPytorch$254k – $305kOn-site · Boston, San Diego and SeattleSenior
Who can apply
On-site · Boston, San Diego and Seattle
Location: Boston, MA / Foster City, CA / San Diego, CA / Seattle, WA
In this role, you will...
Design, train, and evaluate vision models that understand passengers, objects, and behaviors inside the vehicle, across detection, pose, and action recognition.
Independently own a model or feature area end to end: curate data, define labeling guidelines, train and evaluate models, and take them through deployment.
Set the quality bar for the models you own, using rigorous evaluation and error analysis so they are reliable enough to support live operations.
Deploy models under real-time and compute constraints shared with the rest of the onboard perception system.
Partner across product, operations, safety, and vehicle software to turn requirements into model specifications, metrics, and validation plans.
Qualifications
MS or PhD in Computer Science, Machine Learning, Computer Vision, or a related field with 6+ years of industry experience (PhD with 3+ years).
Deep expertise in computer vision and deep learning for image and video understanding: object detection, pose and action recognition, and fine-grained classification, with hands-on experience with modern architectures (convolutional and vision-transformer detectors, and self-supervised or foundation-model backbones such as DINOv3).
Strong proficiency in Python and PyTorch for model design, training, and evaluation.
Experience with vision-language models or foundation-model backbones, and with auto-labeling or synthetic data for long-tail and rare classes.
Track record of owning the full ML lifecycle on real-world data: dataset curation and labeling guidelines, training, metric design (precision, recall, and false-positive rate at an operating point), and error analysis.
Bonus Qualifications
Experience with in-cabin or occupant monitoring (occupant or driver monitoring systems), ideally including multi-camera or fisheye imagery and low-light conditions.