AI-Powered OPD Insurance Claim Adjudication System

Demo: Watch Demo

Overview

An AI-powered insurance claim adjudication platform that automates OPD claim processing using Large Language Models, rule-based policy validation, confidence scoring, and manual review workflows.

The system extracts information from prescriptions, medical bills, pharmacy bills, and diagnostic reports, evaluates them against insurance policies, and generates automated claim decisions.

---

Key Features

AI Document Processing

• Upload prescriptions, medical bills, pharmacy bills, and diagnostic reports
• PDF and image support
• Gemini 2.5 Flash powered extraction
• Structured JSON generation
• Confidence scoring

Automated Claim Adjudication

• Policy rule validation
• Coverage verification
• Medical necessity checks
• Limit validation
• Manual review routing
• Partial approval support

Claim Management

• Claim history tracking
• Detailed audit trail
• Decision explanations
• Confidence score reporting

Administration

• Admin policy configuration dashboard
• Appeals and re-review workflow
• Automated policy management
• Analytics dashboard

---

Technology Stack

Frontend

• Next.js
• TypeScript
• Tailwind CSS

Backend

• FastAPI
• SQLAlchemy 2.0
• Pydantic

Database

• PostgreSQL
• Supabase

Storage

• Supabase Storage

Artificial Intelligence

• Gemini 2.5 Flash

---

System Architecture

                    ┌────────────────────┐
                    │      Next.js       │
                    │    Frontend UI     │
                    └─────────┬──────────┘
                              │
                              ▼
                    ┌────────────────────┐
                    │      FastAPI       │
                    │      Backend       │
                    └─────────┬──────────┘
                              │
          ┌───────────────────┼───────────────────┐
          │                   │                   │
          ▼                   ▼                   ▼

 ┌────────────────┐  ┌────────────────┐  ┌────────────────┐
 │ Supabase       │  │ Gemini 2.5     │  │ Rule Engine    │
 │ Storage        │  │ Flash          │  │                │
 └────────┬───────┘  └────────┬───────┘  └────────┬───────┘
          │                   │                   │
          └───────────────────┼───────────────────┘
                              │
                              ▼
                    ┌────────────────────┐
                    │ PostgreSQL         │
                    │ Supabase Database  │
                    └────────────────────┘

---

Claim Processing Flow

Upload Documents
       │
       ▼
Store in Supabase Storage
       │
       ▼
Gemini Extraction
       │
       ▼
Structured JSON
       │
       ▼
Rule Engine Evaluation
       │
       ▼
Decision Generation
       │
       ▼
Store Results
       │
       ▼
Display Outcome

---

Database Schema

claims

Field	Description
id	Claim ID
status	APPROVED / REJECTED / PARTIAL_APPROVAL / MANUAL_REVIEW
claimed_amount	Submitted claim amount
approved_amount	Final approved amount
confidence_score	AI confidence score
decision_reason	Decision explanation
created_at	Creation timestamp

documents

Field	Description
id	Document ID
claim_id	Associated claim
document_type	Prescription / Bill / Report
storage_url	Supabase Storage URL

extracted_data

Field	Description
id	Record ID
claim_id	Associated claim
extracted_json	Gemini extracted data

---

Project Structure

insurance-ai-adjudicator/

├── frontend/
│
│   ├── app/
│   │   ├── page.tsx
│   │   ├── upload/
│   │   ├── claims/
│   │   └── claims/[id]/
│   │
│   ├── components/
│   │   ├── UploadForm.tsx
│   │   ├── ClaimCard.tsx
│   │   ├── DecisionCard.tsx
│   │   └── LoadingSpinner.tsx
│   │
│   ├── lib/
│   └── types/
│
├── backend/
│
│   ├── app/
│   │   ├── routes/
│   │   ├── services/
│   │   ├── models/
│   │   ├── schemas/
│   │   ├── utils/
│   │   └── core/
│   │
│   ├── requirements.txt
│   └── Dockerfile
│
└── README.md

---

Implemented Adjudication Rules

Approval Conditions

A claim is approved only if all validations pass successfully.

Rejection Rules

PER_CLAIM_EXCEEDED

Reject when claim amount exceeds ₹5000.

BELOW_MIN_AMOUNT

Reject when claim amount is below ₹500.

DOCTOR_REG_INVALID

Reject when doctor registration is missing or invalid.

Valid Examples:

MH/12345/2020
KA/56789/2019
DL/99999/2022

MISSING_DOCUMENTS

Required documents:

• Prescription
• Medical Bill

NON_COVERED_TREATMENT

Examples:

• Teeth Whitening
• Hair Transplant
• Weight Loss Treatment

LATE_SUBMISSION

Reject when submitted more than 30 days after treatment.

DATE_MISMATCH

Reject when treatment dates across documents do not match.

---

Manual Review Rules

LOW_CONFIDENCE

Claims are routed to manual review when:

Confidence Score < 0.80

HIGH_VALUE_CLAIM

Claims are routed to manual review when:

Claim Amount > ₹25,000

---

Partial Approval

Example:

Root Canal          ₹2200
Teeth Whitening     ₹1800

Decision:

PARTIAL_APPROVAL

Approved Amount: ₹2200
Rejected Amount: ₹1800

---

Confidence Scoring

confidence =
(extraction_confidence × 0.7)
+
(document_completeness × 0.3)

Claims with confidence below 0.80 are automatically routed for manual review.

---

API Endpoints

Upload Documents

POST /api/claims/upload

Process Claim

POST /api/claims/process

List Claims

GET /api/claims

Claim Details

GET /api/claims/{id}

---

Local Development

Backend

cd backend

python -m venv venv

venv\Scripts\activate

pip install -r requirements.txt

uvicorn app.main:app --reload

Backend:

http://localhost:8000

Swagger Documentation:

http://localhost:8000/docs

---

Frontend

cd frontend

npm install

npm run dev

Frontend:

http://localhost:3000

---

Environment Variables

Backend

DATABASE_URL=

SUPABASE_URL=

SUPABASE_KEY=

GEMINI_API_KEY=

---

Sandbox Mode

The application supports a sandbox mode for development without API keys.

Sandbox mode:

• Saves files locally
• Generates mock extraction responses
• Simulates adjudication results
• Enables end-to-end testing without external services

---

Implemented Advanced Features

• AI-powered document extraction
• Confidence score based decision making
• Manual review workflow
• Partial approval workflow
• Appeals and re-review workflow
• Admin policy configuration dashboard
• Analytics dashboard
• Automated policy management
• Detailed audit trail
• Claim history tracking

---

Future Enhancements

• Fraud Detection Engine
• Multi-Policy Support

---

Author

Vedant Kasar

Built using FastAPI, Next.js, Gemini 2.5 Flash, PostgreSQL, Supabase Storage, and SQLAlchemy.