anomaly detection · time series · stream processing · data engineering
Telemetry Sentinel
Detecting speed losses in Formula 1. Combines telemetry and race context to distinguish a car problem from a pit stop or yellow flags.
- Role
- Design, implementation and evaluation
- Period
- September - October 2026
- Status
- research
- Visibility
- Case study
The problem
A speed loss can mean a mechanical problem, a pit stop or a yellow flag. I built a detector that compares each car with its recent laps and uses race context to filter expected slowdowns.
How it works
Events are processed in their publication order. A robust reference is computed from previous clean laps for each part of the circuit. A persistent unexplained deficit produces an alert and an evidence snapshot.
What I built
Data preparation, detection, context rules, evaluation and a race replay interface. Experiments compare a robust reference with a quantile model.
Experimental evidence
The current report includes Monza, Japan and Singapore as test races, Spain for validation and Bahrain for development. Real episodes are scarce and Singapore contains false alerts. The public website lets readers explore saved runs and their provenance.