Titan AI Business Analytics Copilot - POC

This is a Proof of Concept (POC) demonstrating how the Model Context Protocol (MCP) can be used at Titan Watches to build an internal AI Business Analytics Copilot. The system routes natural language queries to backend analytics tools using Google Gemini's function calling.

Architecture

• Frontend: Vanilla JS, HTML, CSS, Chart.js for visualization.
• Backend: FastAPI (Python) serving as the MCP Gateway.
• AI: Google Gemini API for intelligent tool selection (intent mapping without keyword matching).
• Database: SQLite (data.db) acting as a Mock Enterprise Data Warehouse.

Files

• database.py: Schema definition and deterministic mock data generation.
• mcp_tools.py: Python implementations of the MCP tools (Sales, Finance, Customer, Service) that query the database.
• main.py: FastAPI server bridging the frontend, Gemini AI, and MCP tools.
• index.html, style.css, script.js: Modern enterprise dashboard.
• requirements.txt: Python dependencies.

Setup & Run Instructions

1. Install Dependencies

   pip install -r requirements.txt
   

2. Initialize Database

   python database.py
   

   (This will create data.db and populate it with realistic mock data)

3. Set API Key (Required for AI)

   export GOOGLE_API_KEY="your-gemini-api-key"
   

   (Note: If the key is not set, the app will run in fallback mock mode)

4. Run the Server

   uvicorn main:app --reload
   

5. Access Dashboard Open your browser and navigate to: http://localhost:8000

Demo Scenarios to Try

Try asking the AI Assistant these queries to see how it routes to different MCP tools:

• "What are our top selling products?" -> Invokes sales_analytics
• "Show me the sales revenue by region." -> Invokes sales_analytics
• "What is our financial summary and net profit?" -> Invokes finance_analytics
• "Give me a breakdown of our customer segments." -> Invokes customer_intelligence
• "Show me the distribution of service request issues." -> Invokes service_analytics