NBFC Loan Management System

AI-powered loan origination, risk assessment, and application management system for Non-Banking Financial Companies.

---

Overview

This system digitises the complete loan lifecycle — from lead capture to automated risk scoring. It replaces manual Excel-based workflows with a structured database, multi-step digital application forms, and a machine learning model that predicts applicant creditworthiness.

Core Workflow

1. Lead Capture — Staff or customers create leads via a digital form
2. Application Submission — Multi-step wizard collects personal, financial, and KYC data
3. Document Processing — PDFs are uploaded and text is extracted automatically via OCR
4. AI Risk Assessment — A Random Forest classifier scores each application (High Risk / Low Risk)
5. Admin Dashboard — NBFC staff review applications in a table or pipeline view, sorted by risk score
6. Status Management — Applications move through the pipeline: New → Under Review → Approved → Rejected → Disbursed

---

Tech Stack

Component	Technology
Frontend	Next.js 16, React 19, Tailwind CSS 4
Backend	Python, Flask, Flask-CORS
Database	Supabase (PostgreSQL)
Auth	Supabase Auth (email/password)
AI/ML	Scikit-learn (Random Forest)
OCR	pdfplumber (PDF text extraction)
Icons	Lucide React

---

Project Structure

NBFCs/
├── backend/
│   ├── app.py              # Flask API — all endpoints
│   ├── train_model.py       # ML training script
│   ├── risk_model.pkl       # Trained Random Forest model
│   ├── migration.sql        # SQL migration for Supabase
│   ├── requirements.txt     # Python dependencies
│   └── .env                 # Environment variables (not committed)
│
├── frontend/
│   ├── src/
│   │   ├── app/
│   │   │   ├── page.tsx                    # Dashboard home with stats
│   │   │   ├── layout.tsx                  # Root layout with navigation
│   │   │   ├── login/page.tsx              # Supabase Auth login
│   │   │   ├── leads/
│   │   │   │   ├── page.tsx                # Leads list with search
│   │   │   │   ├── new/page.tsx            # Create lead form
│   │   │   │   └── [lead_id]/page.tsx      # Lead detail + applicants
│   │   │   └── applications/
│   │   │       ├── page.tsx                # Admin dashboard (table + pipeline)
│   │   │       ├── new/page.tsx            # Multi-step application wizard
│   │   │       └── [application_id]/page.tsx  # Application detail + risk gauge
│   │   ├── components/
│   │   │   ├── AuthNav.tsx                 # Sign in / sign out nav
│   │   │   └── RequireAuth.tsx             # Auth guard wrapper
│   │   └── lib/
│   │       ├── api.ts                      # API client + TypeScript types
│   │       └── supabaseClient.ts           # Supabase client init
│   ├── package.json
│   └── .env.local            # Environment variables (not committed)
│
└── README.md

---

Setup

Prerequisites

• Python 3.10+
• Node.js 18+
• A Supabase project

1. Database Migration

Run the following SQL in Supabase Dashboard → SQL Editor:

-- Add risk assessment columns to application table
ALTER TABLE application ADD COLUMN IF NOT EXISTS risk_score float;
ALTER TABLE application ADD COLUMN IF NOT EXISTS risk_label text DEFAULT 'pending';
ALTER TABLE application ADD COLUMN IF NOT EXISTS app_status text DEFAULT 'New';

-- Document table for uploaded files
CREATE TABLE IF NOT EXISTS document (
  document_id serial PRIMARY KEY,
  application_id int REFERENCES application(application_id),
  applicant_id int REFERENCES applicant(applicant_id),
  doc_type text NOT NULL,
  file_url text NOT NULL,
  ocr_text text,
  ocr_verified boolean DEFAULT false,
  uploaded_at timestamptz DEFAULT now()
);

2. Backend

cd backend
python -m venv venv
source venv/bin/activate    # Windows: venv\Scripts\activate
pip install -r requirements.txt

Create backend/.env:

SUPABASE_URL=https://your-project.supabase.co
SUPABASE_SERVICE_ROLE_KEY=your-service-role-key
PORT=5001

Train the ML model (one-time):

python train_model.py

Start the server:

python app.py

Backend runs at http://localhost:5001

3. Frontend

cd frontend
npm install

Create frontend/.env.local:

NEXT_PUBLIC_API_BASE_URL=http://localhost:5001
NEXT_PUBLIC_SUPABASE_URL=https://your-project.supabase.co
NEXT_PUBLIC_SUPABASE_ANON_KEY=your-anon-key

Start the dev server:

npm run dev

Frontend runs at http://localhost:3000

---

API Endpoints

Public

Method	Path	Description
GET	/api/health	Health check

Authenticated (Bearer token required)

Method	Path	Description
GET	/api/leads	List leads (optional ?q= search)
POST	/api/leads	Create a lead
GET	/api/leads/:id	Get lead by ID
DELETE	/api/leads/:id	Delete a lead
GET	/api/leads/:id/applicants	List applicants for a lead
GET	/api/applicants	List all applicants
POST	/api/applicants	Create an applicant
GET	/api/applications	List all applications
POST	/api/applications	Create an application
GET	/api/applications/:id	Get application by ID
PATCH	/api/applications/:id/status	Update application status
POST	/api/applications/:id/assess	Run AI risk assessment on application
POST	/api/risk-assess	Standalone risk prediction
POST	/api/documents/upload	Upload PDF + OCR extraction
GET	/api/documents/:application_id	List documents for an application
GET	/api/stats	Dashboard summary statistics

---

AI Risk Assessment

The risk engine uses a Random Forest Classifier trained on synthetic NBFC loan data.

Features used:

• monthly_income
• cibil_score
• employment_type (encoded: salaried, self_employed, business, freelancer, retired)
• loan_amount
• loan_tenure
• age
• debt_to_income (derived)
• emi_to_income (derived)

Output:

• risk_score — probability of Low Risk (0.0 to 1.0)
• risk_label — "Low Risk" or "High Risk"

Model performance: 87% accuracy on held-out test set.

The model can be retrained on real historical data by modifying train_model.py and running it again.

---

Features

Lead Management

• Create, search, view, and delete leads
• Linked applicant information

Multi-Step Application Form

• 4-step wizard: Personal Info → Financial Details → Documents → Review
• Client-side validation (age >= 18, CIBIL >= 300, etc.)
• Live AI risk preview before submission

Document Processing

• PDF upload with automatic text extraction
• OCR verification status badges

Admin Dashboard

• Table view — all applications sorted by risk score
• Pipeline view — Kanban-style columns by status
• Inline status updates and AI assessment triggers

Application Detail

• SVG risk score gauge
• Loan details grid
• Document list with extracted text preview

---

Environment Variables

Backend (backend/.env)

Variable	Description
SUPABASE_URL	Your Supabase project URL
SUPABASE_SERVICE_ROLE_KEY	Service role key from Supabase
PORT	Server port (default: 5001)

Frontend (frontend/.env.local)

Variable	Description
NEXT_PUBLIC_API_BASE_URL	Flask backend URL (e.g. http://localhost:5001)
NEXT_PUBLIC_SUPABASE_URL	Your Supabase project URL
NEXT_PUBLIC_SUPABASE_ANON_KEY	Anon/public key from Supabase