Senior Machine Learning Engineer, Vision Models
$311,850–$370,000 year
On-siteSunnyvale, California, United States
Job Summary
Build and fine-tune scene understanding models for Wayve's offline measurement system, adapting on-vehicle architectures and foundation models to assess driving performance across vehicles, markets, and conditions. Diagnose failure modes, define ground truth criteria, and establish automated benchmarks to validate safety cases and steer model iterations. Leverage offline compute advantages for bidirectional temporal context and multi-task learning while measuring your own models rigorously to ensure statistically defensible results. Collaborate with on-vehicle modeling, data curation, and simulation teams to drive accuracy and generalization.
Required Qualifications
- 4+ years in ML engineering, including training and shipping deep learning models in production
- comfortable taking ambiguous modelling problems from scoping through to a working solution
- Hands-on experience training modern computer vision models, including transformer-based and multimodal or VLM architectures for detection, segmentation, classification, or scene understanding, on camera and/or lidar sensor data
- Experience adapting or fine-tuning large pretrained or foundation models, and training shared representations across multiple tasks or objectives (multi-stage or joint training), including real trade-offs across data and losses
- Proficient in Python and ML frameworks (esp. PyTorch), with a solid foundation in software engineering practices and comfort with large-scale training
- Strong ownership: research-literate and pragmatic, able to drive a significant modelling workstream with autonomy, collaborate across teams, and mentor less experienced engineers
- Able to measure your own models: comfortable defining and reading the metrics that show whether a model is genuinely improving
Desired Qualifications
- Experience in 3D scene understanding and representation learning for geometric and semantic perception, including large-scale semantic enrichment of driving scenes
- Experience with offboard or offline modelling: auto-labelling, model distillation, temporal or world models, or other ways of exploiting compute that on-vehicle systems cannot
- Prior experience in autonomous vehicles or robotics with hands-on deployment and closed-loop validation on physical systems
- Experience with fleet-scale data and large-scale distributed training infrastructure
Hiring someone like this?
Get your role in front of qualified candidates on Sorce.