Senior Machine Learning Engineer, Vision Models
Wayve is a London-headquartered embodied-AI company developing and licensing mapless, vehicle-agnostic driving software for assisted, automated, and robotaxi applications.
About Wayve
Wayve Technologies Ltd. develops the Wayve AI Driver, an end-to-end, data-trained software platform that runs on onboard vehicle compute and native sensors. It is designed for OEM integration across L1 driver assistance through L4 automated driving, without HD maps.
Skills
Candidate Availability
Required and preferred rules are kept separate and reflect the wording in the original posting.
About the Role
You will build, train, and fine-tune offline scene-understanding models. You will improve their accuracy across vehicles, geographies, and conditions; define ground truth and correctness criteria; and create automated benchmarks. You will analyse errors, produce statistically defensible evidence, collaborate across sites, and mentor engineers.
Requirements
- 4+ years of machine learning engineering experience
- Experience training and shipping production deep learning models
- Experience training computer vision models for detection, segmentation, classification, or scene understanding
- Experience with transformer-based, multimodal, or vision-language model architectures
- Experience with camera or lidar sensor data
- Experience adapting or fine-tuning pretrained or foundation models
- Experience with multi-task, multi-stage, or joint training
- Python and PyTorch proficiency
- Software engineering and large-scale training experience
- Ability to define and interpret model metrics
- Experience in 3D scene understanding and representation learning
- Experience with offboard or offline modelling
- Experience in autonomous vehicles or robotics
- Experience with fleet-scale data and distributed training infrastructure
Responsibilities
- Build, train, and fine-tune offline scene-understanding models
- Improve model accuracy and generalisation across vehicles, geographies, and conditions
- Diagnose failure modes and address model blind spots
- Use offline compute, temporal context, and joint representation learning
- Benchmark models and set quality bars
- Define ground truth and correctness criteria
- Create statistically defensible evidence for validation pipelines and safety cases
- Collaborate with modelling, evaluation, data curation, and simulation teams
- Mentor engineers and contribute to engineering and modelling practices
Benefits
- Competitive equity package
- Hybrid working arrangement