ShopBot AI

ShopBot AI — MCP-powered E-commerce Support Agent

ShopBot AI is an e-commerce support chatbot I built to explore how far you can push LLMs when they’re connected to real systems like databases and retrieval pipelines.

Instead of relying only on prompt-based answers, the bot can actually:

- check real orders stored in MySQL
- search a knowledge base using embeddings (FAISS)
- decide when it should use tools vs when it should just answer normally

The idea was simple:
a chatbot that doesn’t hallucinate when it matters (like order status).

🔗 Live demo:[https://shopbot-ai-os4z.onrender.com

Most chatbot demos feel impressive at first, but break down quickly when you ask:

- “Where is my order?”
- “What exactly is your refund policy?”
- “Is this product in stock?”

They usually respond confidently… even when they shouldn’t.

I wanted to build something closer to how a real support system should behave:

- if it’s factual → query a database
- if it’s semantic → search knowledge base
- otherwise → let the model respond normally

This pushed me into combining RAG, tool calling, and routing logic in one system.

User message ↓ Flask API ↓ Gemini decides intent (route query) ↓ MCP tool layer ├── MySQL tool → order / customer data └── FAISS tool → FAQ / policy retrieval ↓ ↓ Final response to user