AI-powered lab receptionist built in n8n. Uses an AI Agent with Google Sheets AI Tools (no Workflow Tools/sub-workflows) to check service availability, prices, turnaround times, and clinic info — with memory for multi-turn conversations.
# HealthLab Africa — AI Receptionist 🧪
An AI-powered virtual receptionist for a medical diagnostics laboratory, built entirely in **n8n** using only **AI Agent Tools** — no Workflow Tools or sub-workflows.
This project was built as a learning exercise to understand how n8n's AI Agent node reasons, routes between multiple tools, and uses memory — before moving on to more advanced Workflow Tool patterns.
---
## What it does
The receptionist (named **Ada**) chats with customers and can:
- Check whether a lab test/service is available
- Look up prices
- Look up turnaround times
- Answer general clinic questions (hours, address, phone, email, home sample collection, payment methods)
- Remember the conversation across multiple messages
- Refuse to guess — it only answers using data pulled live from Google Sheets
The Agent itself decides which tool(s) to call based on the customer's message — there is no manual IF/Switch routing logic anywhere in this workflow.
---
## Google Sheets structure
One spreadsheet, four tabs:
| Tab | Columns |
|---|---|
| **Services** | Service ID, Service Name, Category, Available, Description |
| **Price** | Service Name, Price |
| **Turnaround** | Service Name, Turnaround Time |
| **Business** | Question, Answer |
Each AI Tool is scoped to exactly one tab — e.g. the Price Tool can only ever read the Price sheet, which is what enforces the "never mix up prices and availability" rule structurally, not just via prompting.
---
## Key design rules
- The Agent **never answers from memory or guesses** — every factual claim (availability, price, turnaround, business info) must come from a live tool call.
- If a tool finds no matching row, the Agent must explicitly say the service isn't offered — not stay silent or assume.
- The Agent only offers next steps that map to an actual tool (price, turnaround, availability, business info) — it does not invent capabilities like appointment booking or test-preparation guidance, since no tool or da …