Research & Papers

[P] Made an AI FIA Steward to predict penalties during a F1 race

An AI model trained on 1900+ FIA rule PDFs can predict race penalties and cite specific regulations.

Deep Dive

A developer and F1 fan, SoumikS17, has created an AI-powered tool called the 'AI FIA Steward' to navigate the sport's notoriously complex rulebook. The model was trained on a dataset of over 1900 official PDF documents, web-scraped from the Fédération Internationale de l'Automobile (FIA) website, covering all races and regulations from 2019 through 2025. This massive corpus allows the AI to understand the nuanced and often debated rules that govern on-track incidents.

To ensure accuracy and reduce AI hallucinations, the system implements a RAG (retrieval-augmented generation) architecture. When a user describes an incident—such as 'leaving the track to gain an advantage'—the tool searches its database, predicts a likely penalty, and, crucially, cites its sources. It provides the top three most relevant official FIA rule documents with specific PDF citations, enabling users to verify the AI's reasoning against the primary regulatory text. The project is publicly available as a demo on Hugging Face Spaces.

Key Points
  • Trained on over 1900 official FIA rule PDFs from 2019-2025 races
  • Uses RAG architecture to reduce hallucinations and provide source citations
  • Predicts penalties for described incidents and cites top 3 relevant rule documents

Why It Matters

Democratizes understanding of complex sports regulations, providing instant, transparent access to official rules for fans and analysts.