Customer Support AI System

This project showcases a multi-agent customer support system built with Python, OpenAI, and Streamlit. It uses intelligent routing to handle refunds, product FAQs, and general inquiries, simulating how a chatbot manages customer interactions in e-commerce.

Live Demo

View Demo | View YouTube video

Features

Multi-Agent Architecture

• Triage (classifier) Agent: Intelligently classifies and routes customer queries (query classification, intent detection, routing decisions)
• Refund (specialized) Agent: Handles return requests with transaction verification (transaction verification, refund processing, customer validation)
• FAQ (specialized) Agent: Answers product and policy questions using knowledge base (knowledge base search, answer synthesis, confidence scoring)

Why Multi-Agent Architecture?

• Separation of Concerns: Each agent specializes in specific domains
• Scalability: Easy to add new agent types or modify existing ones
• Maintainability: Clear boundaries between different business logic
• Testing: Independent testing of each agent's functionality

Smart Query Processing

• Natural language understanding for intent classification
• Automatic extraction of transaction details (Invoice numbers, Customer IDs)
• Fuzzy matching for FAQ searches with confidence scoring

Session Management

• SQLite database for conversation logging
• Session-based context preservation, query limits, and timeout handling
• Automated cleanup for privacy compliance
• Export capabilities for CRM integration

Caching Strategy

• Response Caching: Reduces API costs for repeated queries
• Session Context: Maintains conversation state
• TTL Management: Prevents stale data issues

Analytics & Reporting

• Downloadable chat logs and refund requests
• Real-time session statistics
• CRM team integration ready

Input Validation & Security

• Sufficient input sanitization
• Transaction verification against database
• Error handling with proper fallbacks

Quick Setup

Prerequisites

• Python 3.12+
• OpenAI API key
• Git

Local Setup

1. Clone the repository

git clone https://github.com/your-username/crm-multi-agents.git
cd crm-multi-agents

2. Run the setup script

python setup.py

This creates the directory structure, generates mock data, and creates .env template.

3. Install dependencies

pip install -r requirements.txt

4. Configure API key Edit .env file:

OPENAI_API_KEY=your-openai-api-key-here
OPENAI_MODEL=gpt-3.5-turbo    # Cost-effective for demo
MAX_TOKENS=150                # Control API costs
TEMPERATURE=0.1              # Consistent responses

MAX_QUERIES_PER_SESSION=30    # Rate limiting
SESSION_TIMEOUT=1800          # 30 minutes
CACHE_TTL=300                # 5-minute response cache

5. Generate mock data (if not done by setup.py)

python create_mock_data.py

6. Launch the application

streamlit run app.py

The app will be available at http://localhost:8501

Docker Setup

# Clone and setup
git clone https://github.com/your-username/crm-multi-agents.git
cd crm-multi-agents

# Create .env with your API key
echo "OPENAI_API_KEY=your-key-here" > .env

# Build and run
docker build -t crm-support-ai .
docker run -p 8501:8501 --env-file .env crm-support-ai

Usage

Sample Interactions

Refund Requests:

User: "I want to return my wireless headphones from order INV1001, customer CUST267"
System: Verifies transaction → Processes refund → Logs request

Product FAQ:

User: "What are the dimensions of PRD001?"
System: Searches knowledge base → Returns specifications → Logs interaction

General Inquiries:

User: "How can I track my order?"
System: Provides tracking information → Offers additional help

Project Structure

crm-multi-agents/
├── agents/
│   ├── __init__.py
│   ├── base_agent.py       # Abstract base class for all agents
│   ├── triage_agent.py     # Query classification and routing
│   ├── refund_agent.py     # Refund request processing
│   └── faq_agent.py        # FAQ and product information
├── utils/
│   ├── __init__.py
│   ├── database.py         # SQLite session management
│   ├── validators.py       # Input validation and sanitization
│   └── response_schema.py  # Standardized response format
├── data/
│   ├── transactions.csv    # Mock transaction database
│   ├── faq.json           # Knowledge base
│   └── session_logs.db    # SQLite session storage
├── config.py              # Configuration management
├── app.py                 # Main Streamlit application
├── create_mock_data.py    # Data generation script
├── setup.py               # Project initialization
├── requirements.txt       # Python dependencies
├── Dockerfile             # Container configuration
└── .env                   # Environment variables (create from template)

Database Structure

Generated by create_mock_data.py execution.

Transaction Database Schema

InvoiceNo,StockCode,Description,Quantity,InvoiceDate,UnitPrice,CustomerID
INV1001,PRD008,Gaming Keyboard,3,2025-07-17,89.99,CUST267

FAQ Knowledge Base Structure

{
  "records": [
    {
      "id": 1,
      "category": "return_policy",
      "question": "What is the return policy?",
      "answer": "Items can be returned within 30 days...",
      "keywords": ["return", "policy", "refund", "30 days"]
    }
  ]
}

Future Enhancements

• Live Chat Integration: Real-time customer support
• API Endpoints: RESTful API for external integrations
• Admin Dashboard: CRM team management interface
• Advanced Analytics: ML-powered conversation and traffic insights

Contributing

1. Fork or star the repository
2. Create a feature branch: git checkout -b feature/your-feature and make your changes
3. Commit your changes: git commit -am 'Add some feature'
4. Push to the branch: git push origin feature/your-feature
5. Submit a pull request

Acknowledgments

• OpenAI for providing the GPT models
• Streamlit for the web framework
• The open-source community for inspiration and tools

---