NexSettle - AI-Powered Insurance Claims Automation Platform

Tech Stack: Django · LangGraph · CrewAI · Gemini 2.5 Flash · Tesseract OCR · MongoDB · Vanilla HTML/CSS/JS

---

Table of Contents

1. Project Overview
2. Architecture
3. Project Structure
4. Prerequisites
5. Environment Variables
6. Quick Start (Windows)
7. Manual Setup
8. Docker Setup
9. One-command Bootstrap
10. API Reference
11. Frontend Pages
12. MongoDB Collections
13. AI Pipeline
14. Credentials (Default)
15. Development Notes
16. CI/CD (GitHub Actions)
17. Render Deployment
18. License

---

Project Overview

NexSettle automates the entire insurance claim lifecycle:

| Step | Technology | Description | |------|-----------|-------------| | Document Upload | Django REST | PDF/PNG/JPG/JPEG multipart upload | | OCR | Tesseract OCR + pdf2image | Text extraction from scanned docs | | Classification | Keyword Detection | Identifies 7+ document types | | Data Extraction | Gemini 2.5 Flash | Structured JSON from document text | | Fraud Detection | Rule-based + AI | Cross-doc validation, format checks | | Policy Verification | MongoDB lookup | Aadhaar/PAN cross-check | | Claim Estimation | Business rules | Natural/Accidental death payout | | PDF Report | ReportLab | Auto-generated settlement report | | Storage | MongoDB | All data in nexsettle_db |

---

Architecture

Architecture diagram file: View SVG

UI Flow

flowchart TD
    A[User Opens /] --> B{Role Action}
    B -->|Register| C[register.html]
    C --> D[OTP Sent]
    D --> E[verify-otp.html]
    E --> F[dashboard.html]
    B -->|Login| G[login.html]
    G --> F
    B -->|Agent Login| H[agent-login.html]
    H --> I[agent-dashboard.html]
    B -->|Admin Login| J[admin-login.html]
    J --> K[admin-dashboard.html]

Agent Flow

flowchart TD
    A[Claim Files Uploaded] --> B[Agentic Orchestrator]
    B --> C[DocumentClassifier Agent]
    C --> D[OCRProcessor Agent]
    D --> E[DataExtractor Agent]
    E --> F[FraudDetector Agent]
    F --> G{Fraud?}
    G -->|Yes| H[Flag Claim + Write fraud_logs]
    G -->|No| I[PolicyVerifier Agent]
    I --> J[ClaimEstimator Agent]
    J --> K[ReportGenerator Agent]
    K --> L[MongoDBStorage Agent]

Backend Flow

flowchart TD
    A[Frontend Request] --> B[Django URL Router]
    B --> C[App View]
    C --> D{Endpoint Type}
    D -->|Auth| E[apps/authentication/views.py]
    D -->|Claim APIs| F[apps/claims/views.py]
    D -->|Pipeline| G[apps/ai_pipeline/views.py]
    G --> H[LangGraph or CrewAI Orchestration]
    H --> I[MongoDB via db/mongo_client.py]
    H --> J[ReportLab PDF Generation]
    I --> K[JSON Response]
    J --> K

Full Project Flow

flowchart LR
    A[User Uploads Claim Documents] --> B[Frontend HTML/CSS/JS]
    B --> C[Django REST API]
    C --> D[Document Intake + Validation]
    D --> E[OCR Engine: Tesseract/pdf2image]
    E --> F[Classification + Extraction]
    F --> G[Gemini 2.5 Flash]
    G --> H[Fraud + Policy Verification]
    H --> I[Claim Estimation]
    I --> J[PDF Report Generation]
    J --> K[MongoDB: nexsettle_db]
    K --> L[Agent/Admin Review]
    L --> M[Settlement Decision]

---

Project Structure

NexSettle_Project/
|-- .github/
|   |-- workflows/
|   |   |-- ci.yml                          # GitHub Actions CI (install, check, smoke tests)
|   |   `-- deploy-render.yml               # GitHub Actions deploy trigger for Render
|-- docs/
|   `-- architecture.svg                    # High-level architecture diagram (linked above)
|-- backend/
|   |-- apps/                               # Django app modules (API domains)
|   |   |-- admins/
|   |   |   |-- apps.py                     # Django app config for admin module
|   |   |   |-- urls.py                     # Admin API routes
|   |   |   `-- views.py                    # Admin auth/dashboard/approval handlers
|   |   |-- agents/
|   |   |   |-- apps.py                     # Django app config for agent module
|   |   |   |-- urls.py                     # Agent API routes
|   |   |   `-- views.py                    # Agent login/review handlers
|   |   |-- ai_pipeline/
|   |   |   |-- apps.py                     # Django app config for AI pipeline
|   |   |   |-- claim_estimator.py          # Claim amount estimation logic
|   |   |   |-- crew_pipeline.py            # CrewAI orchestration flow
|   |   |   |-- data_extractor.py           # Structured field extraction logic
|   |   |   |-- document_classifier.py      # Document type classifier
|   |   |   |-- fraud_detector.py           # Fraud signal checks
|   |   |   |-- pipeline.py                 # LangGraph/LangChain orchestration flow
|   |   |   |-- policy_verifier.py          # MongoDB policy cross-verification
|   |   |   |-- urls.py                     # Pipeline API routes
|   |   |   `-- views.py                    # Upload/process endpoints
|   |   |-- authentication/
|   |   |   |-- apps.py                     # Django app config for auth module
|   |   |   |-- urls.py                     # Register/login/OTP API routes
|   |   |   `-- views.py                    # Auth and OTP handlers
|   |   |-- claims/
|   |   |   |-- apps.py                     # Django app config for claims module
|   |   |   |-- urls.py                     # Claim listing/detail/status routes
|   |   |   `-- views.py                    # Claim API logic
|   |   |-- documents/
|   |   |   |-- apps.py                     # Django app config for documents module
|   |   |   |-- urls.py                     # Document endpoints
|   |   |   `-- views.py                    # Document upload/access handlers
|   |   |-- fraud_detection/
|   |   |   |-- apps.py                     # Django app config for fraud module
|   |   |   |-- urls.py                     # Fraud log/check routes
|   |   |   `-- views.py                    # Fraud API handlers
|   |   `-- reports/
|   |       |-- apps.py                     # Django app config for reports module
|   |       |-- report_generator.py         # ReportLab PDF generation logic
|   |       |-- urls.py                     # Report API routes
|   |       `-- views.py                    # Report download/generate handlers
|   |-- db/
|   |   |-- mongo_client.py                 # Central MongoDB client + DB getter
|   |   `-- __init__.py                     # DB package marker
|   |-- management/
|   |   `-- management/
|   |       `-- commands/                   # Custom Django management commands
|   |           |-- backfill_user_ids.py    # Backfills missing user_id values
|   |           |-- bootstrap_project.py    # One-command setup/bootstrap
|   |           |-- seed_admin.py           # Seeds default admin user
|   |           |-- seed_policy_holders.py  # Seeds policy holder sample data
|   |           |-- setup_mongodb.py        # Creates DB collections + indexes
|   |           `-- smoke_test_flow.py      # End-to-end flow validation command
|   |-- media/
|   |   |-- claims/                         # Uploaded claim files (runtime generated)
|   |   `-- reports/                        # Generated PDF reports
|   |-- nexsettle/
|   |   |-- frontend_views.py               # Serves frontend files via Django
|   |   |-- settings.py                     # Django settings + env configuration
|   |   |-- urls.py                         # Root URL router
|   |   |-- wsgi.py                         # WSGI entrypoint
|   |   `-- __init__.py                     # Project package marker
|   |-- scripts/
|   |   `-- seed_admin.py                   # Script-level admin seeding utility
|   |-- utils/
|   |   |-- id_generators.py                # Claim/user ID generation helpers
|   |   |-- jwt_utils.py                    # JWT create/verify helpers
|   |   |-- masking.py                      # Aadhaar/PAN/account masking utilities
|   |   |-- ocr.py                          # OCR helper wrappers (Tesseract/PDF)
|   |   |-- validators.py                   # Validation helpers/regex checks
|   |   `-- __init__.py                     # Utils package marker
|   |-- .env                                # Local environment values (never commit real secrets)
|   |-- .env.example                        # Environment template for setup/deploy
|   |-- Dockerfile                          # Backend image build instructions
|   |-- manage.py                           # Django management entrypoint
|   |-- Procfile                            # Render process command definition
|   |-- requirements.txt                    # Core Python dependencies
|   |-- requirements-crewai-tools.txt       # Optional CrewAI tools dependency set
|   |-- runtime.txt                         # Render Python runtime version
|   `-- start_render.sh                     # Render startup script (collect/check/run)
|-- frontend/
|   |-- index.html                          # Landing page
|   |-- css/
|   |   `-- main.css                        # Global frontend styling
|   |-- js/
|   |   |-- api.js                          # API client functions (fetch wrappers)
|   |   `-- ui.js                           # Frontend UI behaviors and interactions
|   |-- pages/
|   |   |-- admin-dashboard.html            # Admin dashboard UI
|   |   |-- admin-login.html                # Admin login page
|   |   |-- agent-dashboard.html            # Agent dashboard UI
|   |   |-- agent-login.html                # Agent login page
|   |   |-- dashboard.html                  # Claimant dashboard page
|   |   |-- login.html                      # Claimant login page
|   |   |-- register.html                   # Claimant registration page
|   |   `-- verify-otp.html                 # OTP verification page
|   `-- assets/                             # Static assets (images/icons, optional)
|-- .gitignore                              # Git ignore rules
|-- docker-compose.yml                      # Local Docker services (backend + MongoDB)
|-- README.md                               # Project documentation
|-- render.yaml                             # Render Blueprint configuration
|-- render_readme.md                        # Detailed Render deployment guide
|-- run.bat                                 # Windows quick-start runner
|-- seed_admin.bat                          # Windows helper to seed admin
`-- structure.txt                           # Generated tree reference file

---

Prerequisites

| Requirement | Version | Download | |------------|---------|----------| | Python | 3.12+ | https://python.org | | MongoDB | 7.0+ | https://mongodb.com/try/download/community | | Tesseract OCR | 5.x | https://github.com/UB-Mannheim/tesseract/wiki (Windows) | | Poppler | latest | https://github.com/oschwartz10612/poppler-windows (Windows) | | Git | any | https://git-scm.com |

Windows Tesseract: After installing, Tesseract is typically at: C:\Program Files\Tesseract-OCR\tesseract.exe This path is already pre-configured in backend\.env.

Windows Poppler: After downloading, add the bin/ folder to your system PATH so pdf2image can find pdftoppm.

---

Environment Variables

| Variable | Default | Description | |----------|---------|-------------| | SECRET_KEY | (dev key) | Django secret key | | DEBUG | True | Django debug mode | | MONGO_URI | mongodb://localhost:27017/ | MongoDB connection string | | MONGO_DB_NAME | nexsettle_db | Database name | | JWT_SECRET | (dev key) | JWT signing secret | | GEMINI_API_KEY | "" | Google Gemini API key (required for AI extraction) | | EMAIL_HOST | smtp.gmail.com | SMTP server | | EMAIL_HOST_USER | "" | Gmail address for OTP emails | | EMAIL_HOST_PASSWORD | "" | Gmail App Password | | TESSERACT_CMD | C:\Program Files\Tesseract-OCR\tesseract.exe | Tesseract path (Windows) |

Get a Gemini API key at: https://aistudio.google.com/app/apikey

---

Quick Start (Windows)

# 1. Clone the repo
git clone https://github.com/SANJAI-s0/NexSettle.git
cd NexSettle_Project

# 2. Start MongoDB (if not running as a service)
mongod --dbpath C:\data\db

# 3. Run the backend (auto-creates venv + installs deps)
run.bat

# 4. In a new terminal, seed the default admin account
seed_admin.bat

# 5. Open the app
# Frontend and API are both served from Django
# Open http://localhost:8000

Set your GEMINI_API_KEY in backend\.env before running — otherwise AI extraction will be skipped and regex fallback is used.

---

Manual Setup

1. Create virtual environment

cd backend
python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt

2. Configure environment

copy .env.example .env
# Edit .env and fill in GEMINI_API_KEY, email credentials

3. Create media directories

mkdir media\claims
mkdir media\reports

4. Seed admin account

python scripts\seed_admin.py

5. Start Django

python manage.py runserver

Backend + frontend are available at http://localhost:8000

6. Open app

Open http://localhost:8000 directly in your browser.

CORS Note: Django is configured with CORS_ALLOW_ALL_ORIGINS = True for local development.

---

Docker Setup

# Build and start all services
docker-compose up --build

# Seed admin (in a separate terminal)
docker exec nexsettle_backend python scripts/seed_admin.py

# Stop
docker-compose down

Services:

• nexsettle_backend → Django at http://localhost:8000
• nexsettle_mongo → MongoDB at localhost:27017

---

One-command Bootstrap

For local/project initialization:

cd backend
python manage.py bootstrap_project
python manage.py runserver

---

API Reference

Authentication POST /api/auth/

| Endpoint | Method | Auth | Description | |----------|--------|------|-------------| | /api/auth/register/ | POST | ❌ | Register new claimant | | /api/auth/verify-otp/ | POST | ❌ | Verify email OTP | | /api/auth/resend-otp/ | POST | ❌ | Resend OTP | | /api/auth/login/ | POST | ❌ | Login (returns JWT) | | /api/auth/logout/ | POST | ✅ | Logout | | /api/auth/profile/ | GET | ✅ | Get profile |

AI Pipeline POST /api/pipeline/

| Endpoint | Method | Auth | Description | |----------|--------|------|-------------| | /api/pipeline/process/ | POST | ✅ User | Upload files & run full AI pipeline |

Request: multipart/form-data with one or more files fields
Response: Full extraction JSON with documents, fraud flag, estimation

Claims GET /api/claims/

| Endpoint | Method | Auth | Description | |----------|--------|------|-------------| | /api/claims/list/ | GET | ✅ User | List own claims | | /api/claims/all/ | GET | ✅ Admin | All claims (paginated) | | /api/claims/<id>/ | GET | ✅ User | Get single claim | | /api/claims/<id>/status/ | PATCH | ✅ Agent/Admin | Update claim status |

Reports GET /api/reports/

| Endpoint | Method | Auth | Description | |----------|--------|------|-------------| | /api/reports/<id>/download/ | GET | ✅ | Download PDF report |

Agent POST /api/agent/

| Endpoint | Method | Auth | Description | |----------|--------|------|-------------| | /api/agent/login/ | POST | ❌ | Agent login | | /api/agent/claims/ | GET | ✅ Agent | Pending claims queue | | /api/agent/claims/<id>/review/ | POST | ✅ Agent | Submit review |

Admin POST /api/admin-panel/

| Endpoint | Method | Auth | Description | |----------|--------|------|-------------| | /api/admin-panel/login/ | POST | ❌ | Admin login | | /api/admin-panel/dashboard/ | GET | ✅ Admin | Platform stats | | /api/admin-panel/claims/ | GET | ✅ Admin | All claims with filters | | /api/admin-panel/claims/<id>/approve/ | PATCH | ✅ Admin | Approve claim | | /api/admin-panel/claims/<id>/reject/ | PATCH | ✅ Admin | Reject claim | | /api/admin-panel/agents/create/ | POST | ✅ Admin | Create agent account |

---

Frontend Pages

| File | URL | Role | |------|-----|------| | frontend/index.html | Landing page | Public | | frontend/pages/login.html | Claimant login | Public | | frontend/pages/register.html | Registration + OTP | Public | | frontend/pages/verify-otp.html | Email verification | Public | | frontend/pages/dashboard.html | Claimant dashboard | User | | frontend/pages/agent-login.html | Agent login | Public | | frontend/pages/agent-dashboard.html | Agent claims queue | Agent | | frontend/pages/admin-login.html | Admin login | Public | | frontend/pages/admin-dashboard.html | Full admin panel | Admin |

---

MongoDB Collections

Database: nexsettle_db

| Collection | Purpose | |-----------|---------| | users | Claimant accounts | | otp_verification | OTP codes for email verification | | policy_holder_data | Pre-existing insurance details | | nominee_details | Nominee information | | claims | Processed claims with AI results | | claim_documents | File paths for uploaded documents | | agents | Agent accounts | | admins | Admin accounts | | fraud_logs | Fraud detection audit trail |

---

AI Pipeline

Execution Modes

| Mode | Config | Description | |------|--------|-------------| | LangGraph | AI_ORCHESTRATOR=langgraph | Graph-based deterministic node execution | | CrewAI (Agentic) | AI_ORCHESTRATOR=crewai + USE_CREW_AI=True | Multi-agent role-based orchestration | | Gemini extraction | USE_GEMINI=True | LLM-assisted structured extraction | | Regex fallback | USE_GEMINI=False or API failure | Deterministic extraction fallback |

End-to-End Pipeline Graph

flowchart TD
    A[Upload Documents] --> B[Input Validator]
    B --> C[Format + Type Gate]
    C --> D[OCR Processor]
    D --> E[Document Classifier]
    E --> F[Data Extractor]
    F --> G[Identity Normalizer]
    G --> H[Fraud Detector]
    H --> I[Policy Verifier]
    I --> J[Claim Estimator]
    J --> K[Report Generator]
    K --> L[Mask Sensitive Fields]
    L --> M[MongoDB Storage]
    M --> N[Single Structured JSON Response]

Stage Responsibilities

| Stage | What It Does | Key Output | |------|---------------|------------| | Input Validator | Validates file presence, type, and claim payload | Accepted/rejected upload | | Format + Type Gate | Applies strict rules (for example death certificate must be image/PDF) | invalid_document when violated | | OCR Processor | Runs OCR for images/scanned PDFs, reads text from text/PDF directly when possible | Extracted text + OCR confidence | | Document Classifier | Detects document type using keywords/layout hints | document_type per file | | Data Extractor | Extracts only required fields (no hallucination) | extracted_data with null for missing fields | | Identity Normalizer | Normalizes Aadhaar/PAN/date formats | Clean identifiers and YYYY-MM-DD dates | | Fraud Detector | Checks mismatches, invalid IDs, suspicious patterns, date inconsistencies | fraud_flag + log signals | | Policy Verifier | Matches extracted identities against policy_holder_data | Verification result | | Claim Estimator | Computes estimated payout using claim type + fraud state | estimated_claim_amount | | Report Generator | Builds PDF settlement report | Report file path | | Mask + Storage | Masks Aadhaar/PAN/account before persistence in MongoDB | Stored safe document |

Strict Validation and Status Rules

| Condition | Status | Behavior | |----------|--------|----------| | OCR confidence below threshold (< 0.6) | failed_ocr | Returns re-upload message for clearer document | | Invalid type/format combination | invalid_document | File rejected by rule gate | | Partial extraction across uploaded docs | partial | Returns available fields, missing as null | | All required processing succeeded | success | Full structured JSON response |

Extraction Rules (Enforced)

1. Missing fields are always returned as null.
2. Aadhaar is extracted by regex and normalized to 12 digits only.
3. PAN is extracted by regex and normalized to uppercase pattern.
4. Date fields are normalized to YYYY-MM-DD.
5. Aadhaar/PAN outputs contain only identifier values in extraction payload.
6. No inferred values are added when source text is absent.

Fraud Detection Signals

1. Name mismatch across submitted documents.
2. Invalid Aadhaar format or checksum-like pattern issues.
3. Invalid PAN format.
4. Extremely low OCR confidence.
5. Suspicious formatting or altered-looking text blocks.
6. Date inconsistencies (death date, registration date, policy timelines).

Privacy and MongoDB Storage Policy

1. Full identifiers are used only during processing.
2. Before MongoDB write, sensitive values are masked.
3. Masking examples: Aadhaar: 123412341234 -> ********1234 PAN: ABCDE1234F -> *****1234F Bank A/C: 123456789012 -> XXXXXXXX9012
4. Stored collections include claims, claim_documents, and fraud_logs.

Output Contract

The pipeline returns one structured object with:

1. Top-level status and claim id.
2. Per-document classification + extracted data + confidence.
3. Global fraud_flag.
4. overall_confidence.
5. Ready-to-store payload for nexsettle_db.

Supported Document Types

| Type | Key Fields Extracted | |------|---------------------| | death_certificate | full_name, date_of_death, cause_of_death, certificate_number, issuing_authority | | aadhaar | aadhaar_number (masked to ****1234) | | pan | pan_number (masked to *****4F) | | bank | account_number, ifsc_code, bank_name, account_holder_name | | policy | policy_number, sum_assured, nominee_name, dates | | fir | fir_number, police_station, incident_description | | hospital_record | diagnosis, treatment_summary, cause_of_death | | newspaper_clipping | headline, publication_date, incident_description |

---

Credentials (Default)

Created by running seed_admin.bat or python scripts/seed_admin.py

| Role | Email | Password | |------|-------|----------| | Admin | [email protected] | Admin@123 |

Agents are created by the admin through the Admin Portal → Agent Management.

Claimants self-register at frontend/pages/register.html.

---

Development Notes

CORS & CSRF

• CORS_ALLOW_ALL_ORIGINS = True is set for local dev.
• JWT in Authorization: Bearer <token> header handles auth — no CSRF required.

Fallback Extraction

• If Gemini API key is missing or the call fails, regex-based extraction runs automatically as a fallback.

OCR Confidence Threshold

• If Tesseract confidence < 0.6 → status: "failed_ocr" + user prompt to re-upload.
• PDFs with embedded text bypass Tesseract (confidence = 0.95).

Claim Estimation Rules

• Natural Death: 100% of Sum Assured
• Accidental Death: 200% (double indemnity)
• Fraud Detected: ₹0 payout

Data Masking (MongoDB storage)

• Aadhaar: 123412341234 → ********1234
• PAN: ABCDE1234F → *****1234F
• Bank Account: 123456789012 → XXXXXXXX9012

---

Built with ❤️ using Django · LangGraph · Gemini 2.5 Flash · MongoDB

---

CI/CD (GitHub Actions)

This repo now includes:

• .github/workflows/ci.yml

  • Runs on push/PR
  • Installs dependencies
  • Compiles Python modules
  • Runs Django checks
  • Runs Mongo-backed smoke flow (smoke_test_flow)

• .github/workflows/deploy-render.yml

  • Triggers Render deployment after successful CI on main/master
  • Also supports manual trigger (workflow_dispatch)
  • Requires GitHub secret: RENDER_DEPLOY_HOOK_URL

Required GitHub Secret

Add this in GitHub repo settings:

• RENDER_DEPLOY_HOOK_URL = your Render deploy hook URL

---

Render Deployment

Render deployment is fully documented in:

• Render Deployment Guide

Quick note:

• Deployment method: Render Blueprint via render.yaml
• Health check: /api/health/
• App root: /

---

License

This project is licensed under the MIT License.

• View License

---

rn