OUTLIER.AI | Customer Churn Prediction & Retention System

OUTLIER AI is an advanced, agentic system designed to not only predict customer churn but also autonomously reason about risk factors and generate tailored retention strategies using Retrieval-Augmented Generation (RAG).

Key Features

• Predictive Intelligence: High-accuracy Random Forest classifier with 82.7% ROC-AUC.
• Agentic Reasoning: A multi-stage AI agent that analyzes SHAP-based churn drivers to understand why a customer is at risk.
• RAG-Powered Strategies: Pulls from a curated Tactical Knowledge Base to recommend industry-standard retention plays.
• Dual Flow Architecture:
  • Advanced Agent: Structured strategy reports with reasoning logs and business disclaimers.
  • Llama-3 Integration: Deep synthesis using Llama-3-70b (via Groq) for conversational strategy generation.
• Cyberpunk UI: High-fidelity, glassmorphic dashboard built with Streamlit.

Tech Stack

• Core AI: Scikit-Learn, SHAP, RandomForest
• Agentic Layer: LangChain, Groq (Llama-3), FAISS
• Dashboard: Streamlit, Python-Dotenv
• Embedding Models: HuggingFace (all-MiniLM-L6-v2)

Project Structure

├── models/             # Saved ML model artifacts (rf_model.pkl, etc.)
├── src/                # Core Logic
│   ├── retention_automation.py # Primary Agentic Retention logic
│   ├── retention_agent.py      # Upstream LLM/RAG integration
│   ├── inference.py            # SHAP & Logic orchestration
│   ├── preprocessing.py        # Data cleaning pipeline
│   └── train.py                # Automated training w/ 5-fold CV
├── app.py              # Main "Nexus" Dashboard entrypoint
├── requirements.txt    # Integrated dependencies (LangChain + ML)
├── .env                # API Key configuration
└── README.md

Setup & Installation

1. Environment Configuration

git clone https://github.com/Vegapunk-debug/customer-churn-agentic-retention-system.git
cd customer-churn-agentic-retention-system
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

2. Configure API Keys (Optional but Recommended)

Create a .env file in the root directory:

GROQ_API_KEY=your_groq_api_key_here

Get your key at console.groq.com

3. Run the System

# To train/validate the model
python src/train.py

# To launch the dashboard
streamlit run app.py

Model Insights

Our training pipeline uses 5-fold cross-validation to ensure robustness:

• Accuracy: ~78.8%
• ROC-AUC Score: 82.7%
• Top Feature Contributions: Contract_Month-to-month, TotalCharges, Tenure.

Deployment Note

When deploying to Streamlit Cloud, do not use a .env file. Instead, add your GROQ_API_KEY to the Secrets tab in the Streamlit Cloud Settings.