F1 Intelligence Hub

Overview

Architected a full-stack analytics hub combining FastAPI and Next.js with a scalable 18-table PostgreSQL schema. Built a 1,400+ feature ML pipeline deploying 4 XGBoost/LightGBM models for lap-time forecasting (R² 0.757) and overtake probability (ROC-AUC 0.741). Real-time WebSocket streaming delivers live race updates to concurrent users via a Celery + Redis pub/sub layer.

Key Highlights

• 4 ML models (XGBoost / LightGBM) — R² 0.757 for lap-time forecasting, ROC-AUC 0.741 for overtake prediction
• 1,400+ features engineered from historical race, weather, and telemetry data
• WebSocket streaming for real-time lap updates to concurrent users
• pgvector similarity search for historical race pattern analysis
• Celery + Redis task queue powering async data ingestion pipelines
• Full MLOps: Docker Compose, GitHub Actions CI/CD, MinIO model registry