All projects

AI AUTOMATION / N8N

AI Receptionist

A conversational customer-support and ordering workflow built in n8n. The case study explains the problem, solution, architecture, execution flow, testing, result and lessons learned using the saved project evidence.

n8nAI AgentGroqPostgreSQLTool CallingRAG

01 / PROBLEM

Useful answers, no invented data.

A conversational assistant needs to answer product, order and policy questions using real business information. For order placement, it also needs a controlled path for changing inventory and a confirmation step before the transaction proceeds.

02 / WHY I BUILT IT

From chat to real business actions.

MOTIVATION

I built this project to explore how an AI Agent could move from answering questions to using explicit business tools for real product, policy and order workflows.

02 / SOLUTION

Explicit tools, not model guesswork.

The n8n workflow connects an AI Agent to business tools for product and stock information, order information, policy retrieval and order placement. The transactional logic is delegated to a dedicated Place Order Core workflow.

Chat TriggerAI AgentBusiness ToolsOrder Core
ComponentWhat I builtEvidence
Chat triggerConversational entry point for customer requests.Workflow screenshot
AI AgentAgent configuration with explicit tool-use behavior.Agent screenshot
Business toolsProduct/stock, order, policy and order-placement connections.Workflow screenshot
Order CoreDedicated workflow for the order-placement path.Order Core screenshot
ValidationConfirmation plus successful/insufficient-stock test cases.Execution screenshots

03 / ARCHITECTURE

Conversation, tools and orders kept apart.

The system is structured so the AI Agent decides which explicit tool is needed, while the order workflow owns the inventory-changing operation.

01Chat Trigger
Receives the customer message
02AI Agent
Interprets the request and selects a tool
03Business Tools
Stock, orders, policy and order placement
04Place Order Core
Runs the transactional path
05Response
Returns the actual result
Captured n8n AI Receptionist workflow
Architecture evidence. The saved n8n canvas shows the chat trigger, AI Agent, model, memory and business tools.

04 / WORKFLOW

From a request to a confirmed order.

  1. Customer request: the user asks for product information or an order.
  2. Agent interpretation: the AI Agent selects an explicit business tool.
  3. Information retrieval: product, order or policy information comes from the connected tool.
  4. Confirmation: the assistant restates the requested product, quantity and customer details.
  5. Transaction: after confirmation, the order path is delegated to the Place Order Core workflow.
Captured AI Agent configuration
Agent configuration. The saved configuration defines tool-use and confirmation behavior.

05 / TESTING

Test both the successful path and an edge case.

The project evidence includes a successful order execution and an insufficient-stock scenario. These tests demonstrate that the portfolio is showing behavior, not only a workflow diagram.

07 / CHALLENGES

The parts that needed the most care.

01

Keeping AI grounded in business data

The agent needed explicit tools so product and order information came from workflow data rather than from a static prompt.

02

Separating conversation from transactions

The order-changing path was kept in a dedicated workflow instead of being embedded directly in the conversation logic.

03

Adding a confirmation step

The workflow was designed to confirm the customer's order details before proceeding.

04

Testing an edge case

The saved evidence includes an insufficient-stock test in addition to the successful order path.

08 / RESULT

What the system demonstrates.

Conversational entryCustomer requests enter through chat.
Explicit toolsBusiness information and actions are handled through defined tools.
Order confirmationThe transaction path waits for customer confirmation.
Tested outcomesSuccessful and insufficient-stock behaviors are captured.

09 / LESSONS LEARNED

What I learned while building it.

AI needs boundaries. Connecting the agent to explicit tools is more reliable than expecting the model to know business state.

Architecture matters. Keeping the conversation layer separate from transactional logic makes the system easier to reason about and test.

Evidence matters. A workflow canvas shows design; execution screenshots show behavior.

10 / EVIDENCE

Captured project evidence.

The gallery below keeps the implementation evidence directly with the case study.

Evidence note: Claims on this page are limited to the saved workflow material and screenshots for this project.
Next projectOrder Processing and Fulfillment