AI Dental Reception & Live Queue System

A reception and queue-management system built for a dental clinic client (not named): an AI receptionist handles conversations, while a PostgreSQL-based queue engine decides who is next and what changes when appointments run late.

In a busy clinic, one running-late appointment shifts everything after it. Patients need accurate updates, and reception needs one reliable view of the queue.

Typically a mix of phone calls, a paper or spreadsheet schedule and manual messages to patients when delays occur.

1. Conversation

   AI receptionist workflow in n8n (LangChain agent via OpenRouter, conversation memory in Postgres).

2. Business logic

   Queue recalculation lives in PostgreSQL stored functions, so there is a single source of truth.

3. Dispatch

   The backend and n8n act as thin dispatchers that call the database functions and send messages.

4. Automation

   Separate workflows for delay notifications and a scheduled 15-minute reminder job.

5. Interface

   Next.js dashboards with role-based access (admin, reception, dentist) using JWT roles.

1. A patient message reaches the AI receptionist workflow.
2. The assistant answers or collects details, and calls the booking and queue functions for anything that changes data.
3. When an appointment runs late, the database recalculates the queue and workflows notify affected patients.
4. A scheduled job sends reminders ahead of appointments.
5. Reception and clinicians see the same live queue in their dashboards.

• n8n
• LangChain
• OpenRouter
• PostgreSQL (plpgsql)
• Node.js / Express
• Prisma
• Next.js
• Docker Compose

• Queue logic is in database functions, not in prompts or workflow nodes, so the result is deterministic and testable.
• The AI handles conversation only. It never calculates queue positions.
• Workflows are kept thin so there is one place to change a rule.

• Role-based access separates admin, reception and clinician views.
• Changes to queue state go through database functions rather than free-form AI output.
• No patient data is used in this public description.

• Queue recalculation was developed as database logic so cascade cases (a delay affecting later appointments) can be checked directly.

Outcome: Built and deployed for a dental clinic client. No operational outcome metrics are published.

Limitations: The WhatsApp channel for the AI receptionist is still being end-to-end tested. Healthcare regulatory requirements are not claimed to be met and would be assessed per engagement.