Chaos Engine: I Built an AI That Settles F1 Pit Stop Arguments

Code

https://github.com/dhruvvvgg/Chaos-Engine

How I Built It

The whole thing runs on Google AI Studio's Build mode, using Gemini as the actual reasoning engine behind every "what if."

The part I cared most about getting right was making sure the AI wasn't just generating a vibe-y paragraph. I wanted it to actually reason over real numbers, not make something up that sounded plausible. So instead of asking Gemini to freeform explain a scenario, I feed it a structured JSON block for each intervention, real pre-intervention pace data, the pit loss baseline for that race, degradation curves for the tyre compound being chosen, and the gap to nearby cars, and I have it return a structured verdict: predicted position on exit, estimated gap, a verdict of better/worse/roughly equivalent, and a confidence level. That constraint made a huge difference in how trustworthy the output actually felt.

The "chaos moments" themselves aren't hand picked either. I didn't want to sit there and manually decide which lap of which race was interesting, because that doesn't scale and it's basically just my opinion dressed up as data. So those moments come from statistical detection, things like a driver's lap time deviating sharply from their own rolling average, or a pit stop landing way outside the field's typical stop window for that phase of the race. The chaos score for each race is built from those detected anomalies, not from me deciding a race was dramatic because I remember it being dramatic.

I also went back and forth a lot on the UI. I wanted it to feel like an actual race engineering tool, not a generic dashboard template, so I leaned into a dark telemetry aesthetic, glass card treatments, and real F1 data (pole times, fastest laps, laps completed) pulled from the actual race weekends instead of made up placeholder numbers.

Prize Categories

Best Use of Google AI: Gemini is the reasoning engine behind every "what if" in the app, it takes real race data (pace, pit loss, tyre degradation, traffic gaps) and returns a grounded verdict instead of a generic guess, plus it writes the "why was this race chaotic" line on each race card from the detected anomaly data. The whole thing was also built using Google AI Studio's Build mode.