GenAI Loan Advisory System

An Explainable Generative AI-based system for loan decision support that combines Machine Learning, Explainable AI, and LLMs to provide transparent, user-friendly financial insights.

---

Overview

This project predicts loan approval using a machine learning model and generates human-readable explanations, financial advice, and counter-offers using Generative AI.

The system follows a hybrid architecture:

• ML for decision-making
• GenAI for explanation and advisory (on-demand)

---

Features

• ✅ Loan Approval Prediction (XGBoost)
• 📊 Risk Score Calculation
• 🔍 Explainability using SHAP
• 📚 RAG-based Regulatory Justification (RBI Guidelines)
• 💬 Sentiment-aware responses
• 🤖 LLM-generated:
  • Explanation
  • Financial Advisory
  • Counter-offers (if applicable)
• ⚡ Lazy GenAI (runs only when needed)

---

Architecture

🔹 Decision Layer (Always Runs)

• XGBoost model predicts:
  • Approved / Rejected
  • Risk score

🔹 Generative Layer (On-Demand)

Triggered when:

• User asks for explanation
• OR risk is borderline

Includes:

• SHAP (feature importance)
• RAG (E5 + Qdrant)
• Sentiment Analysis
• LLM (Gemini / Qwen)

---

System Flow

1. User inputs loan details
2. XGBoost predicts decision
3. Show result immediately

If needed: 4. SHAP explains decision
5. RAG retrieves RBI guidelines
6. Sentiment detected
7. LLM generates:

• Explanation
• Advisory
• Counter-offer

---

Tech Stack

Backend

• FastAPI / Flask

Machine Learning

• XGBoost
• SHAP

RAG

• E5 Embedding Model
• Qdrant Vector Database

LLM

• Gemini API (Google AI Studio) / Qwen

---

Dataset

• German Credit Dataset (Kaggle)
• Optional: Synthetic data for augmentation

---

Key Design Principles

• Separation of decision (ML) and explanation (LLM)
• On-demand GenAI (cost efficient)
• Grounded responses using RAG
• Modular architecture

---

📌Example Output

Initial Output: