A modern, AI-assisted motor insurance claim platform built for the Sri Lankan market. Policyholders report accidents, upload photos and documents, and track claims in real time. Garages submit itemised repair estimates. Insurers review AI damage assessments, reconcile garage quotes, detect fraud, and confirm final payouts — all in one place.
| Role | What they do |
|---|---|
| Policyholder | File claims, upload images/documents, view AI damage assessment, track status, receive notifications |
| Garage | Review assigned claims, submit line-item estimates, communicate repair details |
| Admin / Insurer | Review claims, run AI analysis, reconcile garage vs AI estimates, set final claimable value, send messages |
- Damage detection — Gemini analyses uploaded vehicle photos and returns structured damage type, severity, location, and affected parts.
- Repair estimation — LKR-calibrated pricing engine scales parts and labour by vehicle class (bike, three-wheeler, car, van, SUV, lorry, bus, tractor) and premium-make uplift.
- Garage reconciliation — Compares garage line items against the AI estimate and flags overcharges, missed damage, extra items, and labour discrepancies with a divergence score.
- Fraud scoring — Hybrid rule + LLM scoring checks policy recency, duplicate plates, document issues, and incident/damage consistency.
- Controlled part vocabulary — A Sri Lankan part catalogue keeps AI outputs consistent and prices legacy damage rows correctly.
In-app notifications triggered automatically for:
- Document rejection with reason
- New garage estimate submitted
- Final claimable value confirmed or updated
Admins can also send manual messages to policyholders.
- Node.js with Express 5
- TypeScript (strict)
- Prisma ORM 6 with SQLite in development
- Google Gemini API via
@google/generative-ai - JWT authentication, bcryptjs password hashing
- Railway deployment ready (
railway.toml, Nixpacks)
- React 19 + Vite 8
- TypeScript
- Tailwind CSS 4
- React Router 7
- Lucide React icons
Smart-Vehicle-Insurance-Claim-System/
├── backend/
│ ├── prisma/
│ │ └── schema.prisma # Database schema
│ ├── src/
│ │ ├── index.ts # Express app entry
│ │ ├── routes/ # API routes (auth, claims, admin, garage, ...)
│ │ ├── middleware/ # Auth middleware for users, admins, garages
│ │ ├── services/ # Business logic & AI integrations
│ │ └── utils/ # Prisma client, Gemini helpers, upload helpers
│ └── .env # Environment variables
├── frontend/
│ ├── src/
│ │ ├── pages/ # Route pages
│ │ ├── components/ # Shared UI components
│ │ ├── context/ # Auth context
│ │ └── services/ # API clients
│ └── .env # Vite environment variables
├── project-brief.html # One-page project brief
├── railway.toml
└── README.md
- Node.js 22.17+ (recommended)
- npm or yarn
- A Google Gemini API key
git clone https://github.com/<your-username>/Smart-Vehicle-Insurance-Claim-System.git
cd Smart-Vehicle-Insurance-Claim-Systemcd backend && npm install
cd ../frontend && npm installBackend — create backend/.env:
DATABASE_URL="file:./prisma/dev.db"
JWT_SECRET="your-strong-random-secret"
GEMINI_API_KEY="your-gemini-api-key"
PORT=5000Important: Use a strong
JWT_SECRET. The app will not start safely if it is missing.
Frontend — create frontend/.env:
VITE_API_BASE_URL=http://localhost:5000/apicd backend
npx prisma db pushcd backend
npx tsx scripts/seedAdmin.tsThe default admin account is:
| Field | Value |
|---|---|
admin@flashclaim.com |
|
| Password | Admin@FlashClaim1 |
Backend:
cd backend
npm run devFrontend:
cd frontend
npm run devOpen http://localhost:5173 and log in.
The following accounts can be used for testing the different roles in Flash Claim.
Email: admin@flashclaim.com
Password: Admin@FlashClaim1
Email: autofix@gmail.com
Password: 123456
Note: These credentials are intended for development/demo purposes. Do not use these passwords for a production deployment.
For demonstration and testing, you can use the following example claim date:
Example Accident Date: September 7, 2026
Example workflow:
- Register or log in as a Policyholder.
- Add a vehicle and its insurance policy.
- Create a new claim using September 6, 2026 as the accident date.
- Upload the required vehicle damage images and documents.
- Run the AI Damage Analysis.
- Review the detected damage, affected parts, severity, and estimated repair cost.
- Assign the claim to the AutoFix garage.
- Log in using the AutoFix garage account.
- Submit an itemised repair estimate.
- Log in as the Admin / Insurer.
- Review the AI assessment and garage estimate.
- Run Garage Reconciliation to identify price differences, extra items, missing items, and labour discrepancies.
- Review the Fraud Score.
- Set the final claimable value.
- The policyholder receives a notification about the updated claim status/value.
| Area | Endpoint | Description |
|---|---|---|
| Auth | POST /api/auth/register |
Register a policyholder |
| Auth | POST /api/auth/login |
Log in |
| Claims | POST /api/claims |
Create a claim |
| Claims | POST /api/claims/:id/analyze |
Run AI damage analysis |
| Garage | POST /api/garage/claims/:id/estimate |
Submit garage estimate |
| Admin | POST /api/admin/claims/:id/fraud-score |
Re-run fraud scoring |
| Admin | POST /api/admin/claims/:id/reconcile |
Re-run estimate reconciliation |
| Admin | PATCH /api/admin/claims/:id/final-value |
Set final claimable value |
| Admin | POST /api/admin/notifications |
Send message to policyholder |
| Notifications | GET /api/notifications |
List user notifications |
The project is configured for Railway deployment:
railway.tomlsets the backend start command andNIXPACKS_NODE_VERSION.- Add the environment variables in the Railway dashboard.
- Railway will provision the database; update
DATABASE_URLaccordingly.
For production:
- Use a managed PostgreSQL database instead of SQLite.
- Set
NODE_ENV=production. - Ensure
JWT_SECRETis at least 32 random characters. - Serve the frontend build from the backend or deploy it separately (e.g. Vercel).
The app uses a 6-model Gemini cascade with automatic fallback. If a model fails or rate-limits, the next model is tried. Configuration lives in:
backend/src/utils/gemini.ts
Structured output is enforced with responseSchema for damage analysis, fraud signals, garage reconciliation, and estimates.
- ✅ Core claim lifecycle and three-role platform
- ✅ AI damage detection and repair estimation
- ✅ In-app notifications (auto-triggers + admin messages)
- ✅ Fraud scoring
- ✅ Garage estimate reconciliation
- ⏳ Vehicle Part Price Intelligence
- ⏳ AI Accuracy & Automation Improvements
- ⏳ Part tiers (OEM vs aftermarket)
- ⏳ FX pricing for imported parts
This is a solo full-stack learning and portfolio project. Pull requests and suggestions are welcome.
MIT — free to use, modify, and deploy.