AboutGithubLinkedin

Race Replay: the 2024 British Grand Prix on a surveyed Silverstone, with Pit Radar's calls as they happen

Side Project2024

Sports Eclipse

Predicting Formula 1 strategy before the pit wall calls it.

Every Formula 1 race is decided by a handful of calls made under pressure: when to stop, which tyre to fit, whether to gamble on the weather. Sports Eclipse is built to see those calls coming.

Pit Radar: strategy prediction. At the heart of Sports Eclipse is a pit decision model that runs lap by lap, using only information available at that point in the race. It infers how wet the track is from the pace of the whole field, tracks every car's tyre age against expected stint length, and turns it into a live probability that each driver will pit, with the reasoning behind every number.

At the 2024 British Grand Prix, it called the switch to intermediates a lap before the field dived into the pits. It flagged Norris for slicks at 97% a full lap before McLaren brought him in. That lap cost him the race, and Hamilton, who pitted a lap earlier, won.

Race Replay. The full race rebuilt on a surveyed 3D model of Silverstone. All twenty cars run to official timing through the dry start, the rain and the scramble for tyres, and a directed story mode cuts the whole race into a minute of broadcast-style replay.

Lap Duel. Two qualifying laps, driven head to head as ghost cars on their real racing lines. A cinematic chase camera rides with them, the circuit paints in whoever was faster through each of 24 mini-sectors, and a linked telemetry trace shows the exact braking point where time was won.

Strategy. Every stint for every driver, with every Pit Radar prediction set against what the team actually did.

Under the hood

A Python data pipeline ingests official F1 timing, car telemetry and weather through FastF1, derives track wetness from field-wide pace, and feeds a causal strategy model. The front end is React, TypeScript and three.js (React Three Fiber). The 3D circuit is modelled from UK Environment Agency LiDAR survey data and OpenStreetMap, with custom shaders. A frame-exact rendering engine produces broadcast-quality replays on any hardware.

Highlights

  • Predictive strategy model: live pit probabilities for every driver, every lap, using only data available at the time.
  • Explainable: every prediction shows the tyre, wear and weather factors behind it.
  • Official data, end to end: real timing, telemetry and weather, nothing simulated.
  • Tenths made visible: cars positioned precisely enough to separate drivers 0.04 s apart.
  • A surveyed circuit: tracks rebuilt from national LiDAR data.
When2024
Built with
PythonFastF1ReactTypeScriptthree.jsReact Three Fiber
Skills
Predictive ModelingData Pipelines3D RenderingCustom ShadersLiDARData Visualization
PreviousTaccidA simulated hospital that trained AI agents, RL envs
All projects
NextJeopardy GeneratorHost your own Jeopardy night