A production-ready AI restaurant chatbot that answers menu questions from verified restaurant data, recognizes returning customers, takes orders conversationally and automates order logging and confirmation.
The application combines Pinecone RAG with HuggingFace embeddings to retrieve relevant restaurant data before Gemini generates a response. FastAPI orchestrates the conversation, MongoDB stores customers, sessions and orders, and a hybrid summarization system preserves conversation context without continuously sending full chat history. When an order is confirmed, the application automatically saves it, logs it to Google Sheets and sends an HTML confirmation email.
Click Play a Run to follow the system from start to finish.
Sadabahar Restaurant Chatbot is a full-stack conversational AI system built with FastAPI, LangChain and Google Gemini. It can greet returning customers by name, answer menu questions using exact information retrieved from a restaurant PDF, collect orders step by step, save orders to MongoDB and Google Sheets, and send branded HTML confirmation emails.
Restaurant support involves repetitive menu questions and manual order handling. A generic LLM can also hallucinate menu items, prices or policies. The system therefore needed grounded restaurant knowledge, persistent customer recognition, controlled conversation memory and a reliable way to turn a completed chat into real operational actions.
Hybrid 3-phase conversation summarization memory
Pinecone RAG using HuggingFace all-MiniLM-L6-v2 embeddings
Grounded menu answers using restaurant PDF data
Anti-hallucination safeguards with verified restaurant information
Phone-based permanent customer recognition
Returning-customer last-order reorder support
Step-by-step conversational order collection
Automatic order processing after confirmation
MongoDB storage for customers, sessions and orders
Automatic Google Sheets order logging
Branded HTML confirmation emails through Gmail SMTP
Gemini API rotation across up to four API keys
Smart RAG skip logic for simple order fields and confirmations
Two-hour inactive-session expiry with persistent customer identity
Session isolation between customers
The application combines Pinecone RAG with HuggingFace embeddings to retrieve relevant restaurant data before Gemini generates a response. FastAPI orchestrates the conversation, MongoDB stores customers, sessions and orders, and a hybrid summarization system preserves conversation context without continuously sending full chat history. When an order is confirmed, the application automatically saves it, logs it to Google Sheets and sends an HTML confirmation email.
