021 | مشروع أتمتة مطعم | أتمتة خدمة العملاء

Summary

This is the first episode in a series about digitally transforming a restaurant. The presenter focuses on automating customer support and order handling, while emphasizing that automation projects should begin by understanding and standardizing the business’s existing processes.

Key Concepts and Workflow

• Automation vs. optimization: Optimization gives staff better tools to do their jobs; automation aims to run a process with less or no human intervention. A restaurant might retain staff to answer phone calls while automating order recording and follow-up.
• Plan around the business: Before building a workflow, establish clear procedures and prepare reliable information, such as the menu, opening hours, restaurant details, and order-handling rules. Choose tools based on cost and order volume.
• Telegram as the entry point: Customers send questions or orders through a Telegram bot. Staff could also use the same system to enter orders received by phone.
• Intent routing: A router classifies incoming messages into three categories:
  1. General greetings or small talk
  2. Questions about the restaurant or menu
  3. Orders and order-related requests, including checking or cancelling an order
• Knowledge base and RAG: Restaurant and menu documents are uploaded from Google Drive, processed into text, and stored in a Supabase vector database. An AI agent can then retrieve relevant information to answer questions or check product prices.
• Order storage: Google Sheets is used as a simple order database for a small restaurant. Higher-volume businesses may need a faster or more scalable alternative, such as Airtable or a database.
• Order records: The example sheet tracks the order time, items, price, order ID, customer name, phone number, address, and status. Possible statuses include new, in the kitchen, shipped or delivered, and cancelled. Cancelling an order changes its status rather than deleting the record, so cancelled orders can still be counted.
• Order tools: The workflow includes tools to read an order by ID, generate an ID, add an order, cancel an order, and query the restaurant knowledge base. The presenter discusses adding time information to a random ID to reduce the chance of duplicates.
• Kitchen notifications: A separate workflow can respond to new orders and notify the kitchen through Telegram or an API-connected device. This keeps kitchen and delivery processes separate from customer support while allowing them to use the order sheet as a shared data source.

Tutorial and Implementation Notes

The presenter builds the workflow step by step in n8n, using a Telegram message trigger, an AI-based router, a switch with separate branches, and Google Sheets and Supabase tools.

Two design options are discussed: using separate agents that share memory, or using one agent whose system prompt changes dynamically according to the router’s decision. The latter avoids repeating work, but the examples reveal risks involving memory, misclassification, and tools being unavailable on the wrong branch.

Testing shows that greetings and questions can be routed incorrectly. Previous messages in memory can also cause the agent to confuse a new question with an unfinished order. The presenter emphasizes that debugging and testing varied scenarios take substantial time, and suggests using a more capable model or an agent with memory and tools to handle context more reliably.

Cancellation also depends on restaurant-specific rules—for example, whether an order can be cancelled after it reaches the kitchen—so those policies must be agreed upon with the business.

The Telegram bot, knowledge-base setup, and some n8n components are covered in earlier videos. A workflow template is said to be available through the presenter’s Telegram channel.

Main Speaker and Sources

• Karim Nabil — Presenter and instructor; explains the restaurant workflow and demonstrates its implementation.
• iTronics — Mentioned as a sponsor of the series.