ApexFlow: F1 Lap-Time Prediction Platform

ApexFlow is a production-grade, end-to-end MLOps platform designed for real-time Formula 1 lap-time prediction. It leverages professional telemetry, automated retraining, and cloud-native orchestration to deliver race-weekend insights.

🏎️ Overview

This project implements a full ML lifecycle including:

• Data Engineering: DVC-tracked telemetry processing with schema validation.
• Model Engineering: XGBoost/LightGBM with Optuna hyperparameter tuning.
• Automated Retraining: Weekly retraining with warm-start on new race data via Apache Airflow.
• Inference Service: Secure FastAPI service with uncertainty estimation.
• Live Telemetry & Frontend: A React/Vite dashboard for real-time race visualization.
• Observability: Prometheus/Grafana stack for performance & drift tracking.

🛠️ Tech Stack

• Frontend: React 19, Vite, TailwindCSS, Recharts, Lucide-React
• Backend API: Python 3.10, FastAPI, Uvicorn
• Orchestration: Apache Airflow
• Database: PostgreSQL (Supabase / Local)
• Tracking: MLflow & DagsHub
• Data Versioning: DVC
• Monitoring: Prometheus + Grafana + Loki
• Deployment: Docker, Vercel (Frontend), Render (Backend), Supabase (DB)

📁 Repository Structure

.
├── .github/workflows/    # CI/CD pipelines
├── config/               # YAML environment configs
├── dags/                 # Airflow DAGs
├── data/                 # Data samples (versioned by DVC)
├── deploy/               # Cloud Run/K8s manifests
├── docs/                 # Detailed documentation
├── frontend/             # React/Vite Frontend Application
├── monitoring/           # Prometheus/Grafana/Loki configs
├── scripts/              # Utility scripts (Continuous Training, etc.)
├── src/apex_flow/        # Core source code
│   ├── api/              # Prediction API layer
│   ├── data/             # Ingestion & validation
│   ├── modeling/         # Training & versioning
│   ├── monitoring/       # Drift & metrics logic
│   └── orchestration/    # Orchestration logic
└── tests/                # Unit, integration, & quality tests

🚀 Quick Start

Prerequisites

• Docker & Docker Compose
• Node.js 18+ & npm
• Python 3.10+
• DagsHub Account (for MLflow/DVC)

1. Environment Setup

Create a .env file in the root directory:

# Security
DAGSHUB_USER_TOKEN=your_token_here
APEX_API_KEY=race-weekend-key-2026

# Supabase (Optional)
# VITE_SUPABASE_URL=...
# VITE_SUPABASE_ANON_KEY=...

2. Run the Full Stack Locally (Docker)

Launch the API, MLflow, Database, and Monitoring stack:

docker-compose up -d

3. Run Frontend Locally

cd frontend
npm install
npm run dev

Access the dashboard at http://localhost:5173.

4. Continuous Training Pipeline

Trigger the automated retraining pipeline (manually or via Airflow):

# Manual Run
python scripts/continuous_training_pipeline.py

# Airflow
# Enable the 'apexflow_weekly_retraining' DAG

☁️ Deployment Strategy

We recommend a cost-effective, split deployment strategy:

1. Frontend: Deployed on Vercel (connects to frontend/ directory).
2. Backend: Deployed on Render or Railway (runs uvicorn src.apex_flow.api.main:app).
3. Database: Managed Supabase instance (PostgreSQL).
4. Orchestration: Self-hosted Airflow or Cloud Composer (for Weekly Retraining).

Important: Ensure DAGSHUB_USER_TOKEN and APEX_API_KEY are set in your deployment environment variables.

🧰 Scripts

• Continuous Training: python scripts/continuous_training_pipeline.py - Fetches latest race, retrains model, and promotes if better.
• Register Model: python scripts/register_model.py - Manual model registration helper.

📚 Documentation

For deep dives, troubleshooting, and contributing, see:

• Architecture & Data Flow
• Deployment Guide
• Troubleshooting Runbooks

---

Built with ❤️ for F1 Engineers.