Key Notes
Production-level SQL, large-dataset construction and at least three years in data science are essential, alongside real-world experiment design and statistical foundations. The scientist analyses real and simulated driving to define metrics, test product requirements and investigate training or inference bottlenecks for AI engineers. The London role is hybrid with two office days each week and requires accessible communication of analytical findings.
What You'll Work On
- Define and iterate performance metrics that organise AI engineering work and measure progress against product and commercial requirements.
- Design robust experiments and targeted off-road measurements for customer requirements, safety questions and autonomous-driving performance behaviour.
- Investigate model-training and inference bottlenecks, generating and validating technical hypotheses for targeted functionality and measurable performance improvements.
Why This Role Matters
- AI Driver engineering teams use performance metrics and experiments to compare changes, making statistical assumptions and test design material to which model improvements they pursue.
- Customer product requirements span real and simulated driving behaviour; targeted off-road measurements test whether the AI Driver delivers the specified safety and performance.
What They Are Looking For
- Experience: At least three years working in a data-science role is essential.
- Technical: Production-level SQL, data-transformation pipelines and querying or building large datasets are essential.
- Technical: Real-world experiment design, test-statistic evaluation and frequentist statistical foundations are essential.
- Technical: A statistical scripting language and relevant data-science or ML packages, such as Python or R and their libraries, are essential.
- Stakeholder: Accessible visualisation and communication of findings, including evidence of influencing team direction, are essential.
- Preferred: Practical ML, statistical best practices, causal inference, Bayesian methods, distributed computing or startup experience is desirable; PyTorch, Spark and Hadoop are examples.