AI Customer Support Agent Implementation for Restaurant

Overview Customer support for a restaurant involves the same handful of questions asked repeatedly, across all hours of the day. Availability, booking requests, menu details, policies. Answering all of it manually is expensive and unnecessary when the answers are already documented. This project is a fully deployed AI customer support agent with a branded chat interface, persistent memory, a vector knowledge base, and automated email actions for both bookings and human escalations.

How It Works The main workflow starts with a chat message trigger. An AI Agent node handles the conversation, guided by a system prompt that defines its role, what tools it has access to, and how to use them. It connects to a PostgreSQL-backed chat memory node so it retains context across a session, and a PGVector database that stores the restaurant's documents, such as menus, FAQs, and policies, which the agent queries to answer customer questions accurately.

When a customer wants to make a booking, the agent triggers a Gmail node that sends a formatted booking request email to the restaurant. When a situation requires human involvement, a second Gmail node fires a support escalation email. Both are handled automatically based on the agent's judgment.

A separate workflow manages the knowledge base. Staff can upload documents through a form, and the workflow processes and stores them in the vector database so the agent always has access to current information.

Tech Stack Built in n8n using PostgreSQL, PGVector, OpenAI, Gmail, and a publicly deployed chat frontend that is fully customizable including the title, greeting message, and brand styling.

The Result A customer-facing support agent that handles routine inquiries 24/7, sends booking requests without staff involvement, and only escalates when it genuinely needs to.