F1 Pit Window Predictor

An automated data pipeline that ingests telemetry data to predict tire degradation and optimal pit stops in real-time.

The Challenge

Formula 1 strategy relies heavily on understanding tire degradation ("deg"). However, deg is non-linear and affected by track temperature, fuel load, and traffic. The goal was to build a system that could ingest raw lap time data and visualize the "crossover point"—the exact moment a new set of tires becomes faster than the current set.

System Architecture

01. IngestionPython scripts pull telemetry from the OpenF1 API, containerized in Docker for consistency.
02. StorageRaw JSON is processed and normalized into a PostgreSQL database schema designed for time-series queries.
03. OrchestrationApache Airflow (running in Docker Compose) schedules the ingestion jobs and manages dependencies.
04. AnalysisPandas & Scikit-learn models calculate the deg curve and predict the pit window.

Tech Stack

PythonDockerAirflowPostgreSQLPandasMLflow

Key Learnings

Managing Airflow DAGs in a containerized environment.
Handling API rate limits and data normalization.
Designing idempotent data pipelines.

Ingestion Logic

# Example: Dockerized ingestion script snippet

def fetch_telemetry(session_key, driver_number):
    """
    Fetches lap data and normalizes for storage.
    Running inside Docker container 'ingest-worker-01'
    """
    endpoint = f"https://api.openf1.org/v1/laps?session_key={session_key}"
    response = requests.get(endpoint)
    
    if response.status_code == 200:
        data = normalize_laps(response.json())
        db.bulk_insert(data)
        logger.info(f"Ingested {len(data)} laps for Driver {driver_number}")
    else:
        logger.error("API Gateway Timeout")