Drishti: SPANDAN Intelligence Platform 👁️

Drishti is an AI-powered prospect intelligence copilot designed specifically for IDBI Bank branch Relationship Managers (RMs). By synthesizing Core Banking System (CBS) transaction data with external credit behaviors, Drishti identifies high-intent cross-sell and up-sell opportunities, empowering RMs to act precisely when the customer needs it most.

---

🚀 Key Features

• Predictive AI Models: Utilizes LightGBM to rank prospects based on Conversion Probability, Capacity, and Urgency.
• Explainable AI (XAI): Integrated SHAP values translate complex ML predictions into human-readable insights ("Why did the AI recommend this?").
• Real-Time Streaming Alerts: Employs Server-Sent Events (SSE) to push instant notifications to the RM's dashboard when high-intent actions (like a competitor auto loan payoff) are detected.
• Mock Integrations: Built-in simulators for CIBIL credit checks and CBS lead generation endpoints.
• Lightning-Fast Modern Stack:
  • Frontend: Next.js (App Router), React, TailwindCSS
  • Backend: FastAPI, SQLAlchemy, SQLite (for portability)
  • ML Pipeline: Python, Pandas, Scikit-Learn, LightGBM

---

🏗️ Architecture

The application is cleanly decoupled into three main layers:

1. ml/ (Offline Pipeline): Batch processes transaction data, runs feature engineering, and calculates AI propensity scores. Outputs to the database.
2. backend/ (FastAPI Core): Serves the REST API, handles user authentication, connects to SQLite, and pushes real-time SSE notifications.
3. frontend/ (Next.js Dashboard): The beautiful, responsive UI consumed by the Relationship Manager.

---

💻 Running Locally

Option 1: Docker (Recommended for Judges)

The entire stack is containerized for instant, dependency-free execution using Docker Compose.

# Start the API and Frontend Web server
docker compose up -d

# (Optional) Run the offline ML Pipeline to generate new scores
docker compose --profile pipeline up pipeline

The frontend will be available at: http://localhost:3000 The backend API will be available at: http://localhost:8000

Option 2: Native Execution (uv & pnpm)

1. Backend (FastAPI)

cd backend
uv sync
uv run uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload

2. Frontend (Next.js)

cd frontend
pnpm install
pnpm run dev

---

🧪 Testing the Application (Demo Mode)

We have included a seed script to instantly populate the database with high-fidelity dummy data so you can test the UI without waiting for an ML batch run.

1. Seed the Database:

cd backend
uv run python scripts/seed_demo_data.py

2. Log In: Navigate to http://localhost:3000 and use the following credentials:

• Email: demo@idbi.co.in
• Password: demo123

You will immediately see 3 rich prospect mockups (Personal Loan, Auto Refinance, Credit Card Upgrade) complete with SHAP factor explanations!

---

🛠️ Built For IDBI Hackathon

Engineered with ❤️ for the IDBI Hackathon. Fully compliant with modern industry best practices including Global Error Boundaries, Exception Interceptors, and strict type safety.