Bazinga 2.0

Members:

• Sarwadnya Mahajan
• Kanaad Limaye
• Vedant Lavale
• Vishwajeet More

Insurance STP — Intelligent Claim Processing

Automated insurance claim processing with Straight-Through Processing (STP) for Car and Travel claims.

---

Quick Start

1. Clone & Install

cd insurance-stp
pip install -r requirements.txt

2. Configure environment

cp .env.example .env
# Edit .env with your keys:
# - ROBOFLOW_API_KEY + ROBOFLOW_MODEL_ID
# - SMTP credentials (Gmail App Password recommended)
# - TELEGRAM_BOT_TOKEN + TELEGRAM_REVIEWER_CHAT_ID

3. Build policy index (run once)

python scripts/build_policy_index.py

Add your policy PDFs to policy_store/raw/ before running. Without PDFs, it auto-generates sample policy chunks so the system still works.

4. Run the server

uvicorn app.main:app --reload --port 8000

5. Open API docs

http://localhost:8000/docs

---

API Endpoints

Method	Endpoint	Description
POST	/car/submit	Submit a car insurance claim
POST	/travel/submit	Submit a travel insurance claim
GET	/review/queue	View all claims pending manual review
POST	/review/{id}/decision	Reviewer approves or rejects a claim
GET	/health	Health check

---

Car Claim — Form Fields

Field	Type	Description
claimant_name	string	Full name
claimant_email	string	Email for notifications
claimant_telegram_id	string (optional)	Telegram chat ID
policy_number	string	e.g. CAR001
description	string	Description of the incident
reported_damage_type	string	e.g. "front bumper", "windshield"
estimated_amount	float	Claim amount in INR
evidence	file	Photo of damage (JPEG/PNG, taken from camera)

Travel Claim — Form Fields

Field	Type	Description
claimant_name	string	Full name
claimant_email	string	Email for notifications
claimant_telegram_id	string (optional)	Telegram chat ID
policy_number	string	e.g. TRAVEL001
description	string	Description of the claim
trip_origin	string	Departure city
trip_destination	string	Destination city
travel_date	string	YYYY-MM-DD
claim_category	string	flight_delay / baggage_loss / medical / trip_cancellation
estimated_amount	float	Claim amount in INR
evidence	file	Document photo (ticket, bill, etc.)

---

STP Decision Logic

Final Score = 0.4 × Consistency + 0.35 × Evidence + 0.25 × Policy

≥ 0.75  →  AUTO APPROVE  →  Settlement + Email/Telegram notification
0.5–0.75 → MANUAL REVIEW →  Reviewer queue + Claimant notified
< 0.5   →  REJECT        →  Email/Telegram with reason

---

Active Policy Numbers (for testing)

• Car: CAR001, CAR002, POL001, POL002, POL003
• Travel: TRAVEL001, TRAVEL002, POL001, POL002, POL003

---

Project Structure

app/
  api/          # Route handlers (car, travel, review)
  pipeline/
    car/        # Metadata check → Validation → YOLO extractor
    travel/     # Metadata check → Validation → OCR extractor
    shared/     # Consistency + Evidence + Policy scoring → STP decision
  services/     # Settlement + Notifications (email + Telegram)
  db/           # SQLAlchemy models + CRUD
  utils/        # File handling, hashing, logging
ml_models/ocr/  # EasyOCR wrapper + field parser
policy_store/   # Raw PDFs → chunked JSON → FAISS + BM25 indexes
scripts/        # build_policy_index.py, seed_db.py