Insurance Fraud & Priority API

A two-stage ML system that validates claim authenticity, then calculates a priority score.

Tech Stack

• BigQuery, SQL (CTEs), Pandas
• XGBoost (Supervised)
• FastAPI, Uvicorn, Pydantic
• Weights & Biases (W&B)

Business Problem Solved

Insurance fraud detection with 80%+ accuracy & claim prioritization by processing urgency (amount + customer tenure), reducing manual audit overhead and accelerating payouts for genuine claimants.

• Hybrid model approach (supervised + unsupervised fusion)
• 3:1 weighting (fraud is 25% of dataset, 753 legitimate vs 253 fraud)

Scale

• FastAPI handling 500-1000 predictions/sec per instance
• Render/Heroku (render.yaml + Procfile) with auto-scaling
• <50ms per prediction