The Problem

ABC Plumbing & Drain was missing calls after hours, losing potential bookings to competitors who picked up the phone. Support emails were going unanswered for hours. The team had no system to handle high volumes of repetitive enquiries — service pricing, availability checks, appointment bookings — without burning staff time.

The client needed an agent that could handle all inbound communication channels, sound professional, check real calendar availability, book appointments, and only escalate genuine emergencies to human staff.

The Solution

I built Emily — a fully autonomous AI customer support agent running on n8n Cloud, powered by GPT-4o-mini as the LLM with Retell AI handling the voice layer. Emily operates across 3 channels simultaneously with smart routing and escalation logic built in.

Tech Stack

The Main Workflow

Emily’s core workflow handles all 3 channels — voice, email, and chat — routing each through dedicated preparation nodes before passing to the AI agent. The agent uses RAG knowledge retrieval, conversation memory, and tool calls to book appointments, check availability, and log every interaction.

Emily — AI Customer Support Agent: Voice + Email + Chat inputs → routers → AI agent → booking / escalation / response branches

Calendar Integration — Book & Check Availability

Emily has two dedicated sub-workflows for calendar management. One checks real-time availability against Google Calendar before offering time slots, and another books confirmed appointments directly — no double bookings, no manual intervention.

Check Availability — queries Google Calendar in real-time before offering slots to the customer

Book Appointment — creates confirmed Google Calendar events automatically after customer confirms

Knowledge Base — Supabase RAG

A one-time setup workflow uploads all plumbing FAQs, pricing information, and privacy policies into Supabase as vector embeddings using Google Gemini. Emily queries this knowledge base on every interaction — giving accurate, consistent answers every time without hallucinating.

Emily Supabase Upload — FAQ field, pricing field, and privacy policy uploaded as vector embeddings via Google Gemini

LLM Model Abstraction Layer

A dedicated sub-workflow dynamically fetches Emily’s active LLM configuration and routes to either Claude API or Grok API based on the config — normalizing the response into a unified format. This means the LLM can be swapped without touching the main workflow at all.

Emily LLM Model Abstraction Layer — fetches config → routes to Claude or Grok → normalizes response → unified output

How Emily Works

Results & Capabilities