🏦 Bank Customer Churn Prediction with Generative AI

A full ML lifecycle project for predicting customer churn in banking, covering data ingestion, feature engineering, model training, evaluation, deployment, and a real-time prediction interface with AI-driven explanations.

This project also includes a bulk churn detection workflow that can automatically generate and send personalized retention emails using LangChain + Groq, orchestrated with LangGraph for stateful AI workflow management.

---

Project Overview

1️⃣ Data Pipeline (DVC)

• Stages: Ingestion → Transformation → Preparation → Training → Evaluation → Registration
• Produces artifacts that feed the next step
• Ensures reproducibility, traceability, and version control

2️⃣ Model Building

• Handles class imbalance with SMOTE
• Feature selection: RFECV
• Models trained: LightGBM & XGBoost → best model selected
• Hyperparameter tuning & early stopping
• Evaluation metrics: Accuracy, Precision, Recall, F1-score, ROC-AUC
• Artifacts saved: model, selected features, evaluation metrics

3️⃣ MLflow Model Registry

• Tracks experiments & metrics
• Models stored in S3
• Supports versioning, staging, production workflow
• Models can be dynamically loaded for prediction

4️⃣ Deployment

• FastAPI serves predictions via REST API
• Supports dynamic model loading from MLflow
• Dockerized for consistent deployment environments
• CI/CD automated using GitHub Actions

5️⃣ Frontend

• Streamlit interface for user-friendly input
• Calls FastAPI endpoint for predictions
• Displays predicted class & probability
• Generative AI explanation:
  • Uses LangChain + Groq to generate business-friendly explanations
  • Explains why the model predicted churn for each customer
  • Helps non-technical users understand model decisions

6️⃣ Bulk Churn Detection & Retention Emails

• Detect churners in bulk using historical customer data
• Generates personalized retention emails automatically for each churner
• Emails use content from retention offer guide PDF
• Emails are sent via Gmail SMTP
• Orchestrated with LangGraph R8 workflow:
  • usermsg → checks user intent
  • predict_churners → predicts churn using ML model
  • send_email → generates & sends emails
  • other_query → handles non-churn queries
• Streamlit button workflow:
  1. Identify churned customers
  2. Display results in a dataframe
  3. Generate & send personalized emails automatically

7️⃣ Logging & Evaluation

• Comprehensive logging for all stages
• Confusion matrices & metrics tracked in MLflow
• Ensures transparent and reproducible evaluation

---

8️⃣ Retrieval-Augmented Generation (RAG) for Banking FAQ Assistant

An integrated RAG-based chatbot that intelligently answers bank-related FAQs using live website data.
The chatbot:

• Dynamically scrapes official bank FAQ and offer pages using Selenium
• Splits and embeds content using HuggingFace Embeddings (all-MiniLM-L6-v2)
• Stores vectors in a Chroma vector database
• Uses LangChain + LangGraph + ChatGroq to:
  • Classify queries
  • Retrieve relevant content
  • Decide contextual relevance
  • Generate concise, factual, and context-based answers

Technologies

• 🐍 Python, LightGBM, XGBoost – Model development
• 📊 DVC – Data pipeline & reproducibility
• 📈 MLflow – Experiment tracking & model registry
• 🐳 Docker – Containerization
• ⚙️ GitHub Actions – CI/CD deployment
• 🚀 FastAPI – Backend API
• 🖥️ Streamlit – Frontend UI
• 🤖 LangChain + Groq – AI-powered model explanations & email generation
• 📌 LangGraph – Orchestrates AI workflow & stateful decision logic