Autonomous healthcare provider data validation using Agentic AI
Healthcare provider directories are broken:
- 80% contain errors (wrong addresses, phones, credentials)
- 40-60 hours/week wasted on manual verification
- $2.4B annual industry cost
MediGuard AI uses multi-agent architecture to autonomously validate provider data:
Process: Upload → Extract → Verify → Predict → Score → Update
Result: 30 minutes vs 40 hours | 92% accuracy | 87% time savings
- OCR Extraction: GPT-4 Vision + Tesseract for document processing
- Parallel Processing: NPI verification + web scraping simultaneously
- ML Prediction: XGBoost model with SHAP explainability
- Intelligent Split: 85% auto-validated, 15% human review
- Real-time Updates: Automated directory sync and compliance reports
| Metric | Before | After | Gain |
|---|---|---|---|
| Time | 40 hrs/wk | 5 hrs/wk | 87% ↓ |
| Accuracy | 65% | 92% | +27% |
| Cost | $120K/yr | $18K/yr | $102K saved |
Frontend: HTML5, CSS3, Responsive Design
Backend: Python, FastAPI, PostgreSQL
AI/ML: GPT-4 Vision, XGBoost, SHAP
APIs: NPI Registry, Twilio, State Medical Boards
EY Techathon 6.0 - Firstsource Challenge
Theme: Agentic AI for Autonomous Business Processes
Anurag Gupta
B.Sc. Computer Science | Ruia College
GitHub | LinkedIn
MIT License - Feel free to use for educational purposes
⭐ Star this repo if you find it useful!