Description
We build the embodied intelligence that moves real vehicles safely, and the ecosystem a billion machines will run on in the future.
As a Machine Learning Engineer, Data Enrichment, you will tackle computer vision problems spanning road geometry, spatial relationships, road users and driving behaviour.
Your day-to-day:
- Work with internal customers to define scene and scenario attributes, dataset requirements and measurable acceptance criteria for annotations and classifiers
- Design and refine annotation tasks and labelling instructions, using pilots, annotator agreement and expert review to resolve ambiguity and measure quality
- Curate and deliver ground-truth datasets for other teams to train and evaluate models, addressing rare scenarios, class imbalance and coverage across countries and vehicle platforms
- Build and evaluate computer vision models that draw on Wayve's World Foundation Model to identify scene attributes and driving behaviours
- Use error analysis to identify gaps in data, labels and model performance, and test approaches to improve them
What you'll be working on:
- Ground-truth datasets and classifiers that describe road geometry, spatial relationships, road users and driving behaviour
- Integrating classifiers with enrichment infrastructure to run inference across existing and newly collected driving data, making predictions available for corpus analysis and scenario mining
- Improving reusable tools for annotation, training, evaluation and inference
- Documenting dataset and classifier quality, and communicating limitations to colleagues across disciplines
You should apply if:
- You have practical experience developing and evaluating computer vision models for image or video understanding, including interpreting spatial relationships in visual data
- You have strong Python programming skills and experience with a modern machine learning framework such as PyTorch
- You have experience curating training and evaluation datasets and assessing annotation quality, with an understanding of class imbalance, data leakage and representative evaluation
- You have taken machine learning models beyond experimentation into use, with reproducible training, evaluation and inference
- You can investigate model failures and determine whether improvements need better data, clearer labels or changes to the modelling approach
- You can translate an ambiguous customer request into clear technical deliverables and quality criteria, and communicate performance and limitations to colleagues from different disciplines