🏎️ TrackVision: Formula 1 Performance Analytics & Prediction

TrackVision is a production-grade AI-powered Formula 1 analytics dashboard that visualizes driver performance trends and predicts upcoming race results. Built using ML architecture and a streamlined data pipeline, TrackVision ingests real race data, trains predictive models, and serves intelligent insights through a modern API layer.

🌐 Live Stack Overview

Component	Description
FastAPI	Backend API for race stats, visualizations, and predictions
scikit-learn	ML model training and evaluation pipeline
AWS S3	Cloud storage for datasets and trained models
Apache Airflow	Orchestrates weekly model retraining and updates
React + Vite	(Frontend) Visual dashboard for race analysis

---

🔍 Features

• 📈 Visualize performance for any current F1 driver
• 🧠 Predict next race finishing position using ML
• ⚙️ Cloud-first design with S3 data ingestion and model storage
• 🚀 Weekly retraining pipeline via Apache Airflow
• 🔄 RESTful API integration with frontend dashboard

---

📊 Machine Learning

TrackVision uses a Gradient Boosting Regressor trained on sliding windows of recent race data to predict a driver's next finishing position. The model considers:

• Average grid and finish positions
• DNF count and lap completion
• Qualifying rank
• Fastest lap rank
• Constructor gap (team points lead vs. current top team)

The model is optimized using GridSearchCV and evaluated via Mean Absolute Error (MAE).

---

📁 API Endpoints

GET /top-driver

Returns the current top driver in the championship standings.

{ "driver": "Oscar Piastri" }

---

GET /visualize?driver={name}&range={1|5|full}

Returns summary stats and race history for a specific driver.

• range=1 → Last race
• range=5 → Last 5 races
• range=full → Entire season

{
    "driver": "Max Verstappen",
    "season": "2025",
    "range": "1",
    "summary": {
        "total_points": 1.0,
        "best_position": 10,
        "races_completed": 1
    },
    "races": [
        {
            "round": 9,
            "race_name": "Spanish Grand Prix",
            "date": "2025-06-01",
            "position": 10,
            "grid_position": 3,
            "points": 1.0,
            "status": "Finished"
        },
        ...
    ]
}

---

POST /predict

Returns the predicted position for a driver in the next race.

{
  "driver": "oscar piastri",
  "predicted_position": 2,
  "confidence": 98,
  "upcoming_race": {
    "name": "Canadian Grand Prix",
    "circuit": "Circuit Gilles Villeneuve",
    "date": "2025-06-15"
  },
  "recent_finishes": [1, 1, 3, 3, 1],
  "average_finish": 1.8,
  "team_points_lead": 192,
  "qualifying_rank": 1
}

---

⚙️ How It Works

1. Data Ingestion

   • Driver results for the 2025 season are stored in Parquet format in S3.
   • Fetched into memory using boto3 and pandas.

2. Model Training

   • train_model.py extracts features from recent race windows.
   • Uses scikit-learn's GradientBoostingRegressor with hyperparameter tuning.
   • Trained model and constructor encoder are saved to S3 for future inference.

3. API Layer

   • Built using FastAPI for high-speed, async-ready responses.
   • Handles endpoints for data visualization and predictions.
   • Uses lru_cache for S3 efficiency and quick response times.

4. Scheduled Retraining

   • An Apache Airflow DAG runs weekly to refresh the model using new race data.

---

🧪 Local Development

🔧 Requirements

pip install -r requirements.txt

🛠️ Environment Variables

Before running the app, set your environment variables in a .env file or export them manually:

AWS_ACCESS_KEY_ID=
AWS_SECRET_ACCESS_KEY=
AWS_DEFAULT_REGION=

TRACKVISION_BUCKET=
CORS_ORIGINS=

🌍 Run FastAPI App

uvicorn app:app --reload

🧪 Run the Airflow DAG (development)

Ensure Airflow is set up and the DAG is loaded, then trigger manually:

airflow dags trigger trackvision_weekly_model_training

🌍 Run the React web server

npm run dev

---

☁️ Deployment

You can deploy the full TrackVision system as follows:

🐳 Docker Compose (Recommended for Full Stack)

Use the included docker-compose.yml to spin up the backend, Airflow, and any other services.

docker-compose up --build

Ensure your .env is properly configured and included.

🧑‍💻 Backend API (FastAPI)

• Deploy to Railway, Render, or [AWS EC2]
• Ensure the container has access to your .env vars and AWS credentials
• If not using Docker, use Gunicorn + Uvicorn worker to serve FastAPI in production

🛰 Frontend (React)

• Deploy to Vercel or Netlify
• Update CORS_ORIGINS in your backend to match the deployed frontend origin

📅 Airflow

• Run locally with Docker Compose, or
• Deploy to managed services like MWAA (Managed Workflows for Apache Airflow) or Astro
• Schedule the DAG (trackvision_weekly_model_training) to retrain the model weekly

---

📸 Screenshots

---

👨‍💻 Author

Adnan T. — @adnant1