HOSPITALITY AUTOMATION
Restaurant Ordering Automation
A Telegram ordering agent that answers from the restaurant's own menu and delivery rules, then puts a manager in front of the money.
- PROJECT
- Restaurant Ordering Automation
- INDUSTRY
- Hospitality Automation
- STACK
- n8n · AI Agents · OpenRouter · Telegram Bot API · Google Sheets · LLMs
OVERVIEW
The project
Taking orders by message means somebody reading and retyping every one of them. Allzone built an agent on Telegram that answers questions from the restaurant's own menu, delivery and FAQ material, captures the confirmed order in structured form, calculates the total, records it and sends payment instructions. When the customer submits a receipt, the order goes to a manager to approve rather than clearing itself.
CHALLENGE
What made it difficult
A conversational ordering bot has to answer from the restaurant's actual menu and delivery rules. Anything it invents becomes an order the kitchen cannot fill.
Payment is where full automation stops being sensible. Reading a receipt image and deciding that money has arrived is exactly the decision worth keeping with a person.
WHAT WE DID
The work delivered.
- 01
Conversational ordering
An agent on Telegram using menu, business information, delivery and FAQ tools, so its answers come from the restaurant's own material.
- 02
Structured order capture
The confirmed order extracted into a structured record with the total calculated and stored, then payment instructions sent to the customer.
- 03
Receipt handling
Payment receipt images detected and attached to the order, with the order status updated on submission.
- 04
Manager approval
The order and its payment detail sent to a manager on Telegram for review, with the approval flowing back into the order status.
ARCHITECTURE
How it fits together
- Restaurant knowledge exposed to the agent as tools rather than pasted into a prompt
- Structured extraction between the conversation and the record, so a stored order is machine readable
- A spreadsheet as the order record, which keeps the flow auditable with no database to run
- A human approval step on payment, with the customer's status updated automatically once it clears
TECHNOLOGY
The stack
- n8n
- AI Agents
- OpenRouter
- Telegram Bot API
- Google Sheets
- LLMs
OUTCOME
What exists now.
Customers order in a chat they already use, and the order arrives structured rather than as a message to retype
Payment verification stays a human decision, which is the one step here where being wrong costs money
The day's orders and their payment state are readable in one place
NEXT STEP
Tell us what you're building.
Bring us the problem with its real constraints attached. We will tell you what we would build, and what we would not.