Customer Contact n8n Workflow

A centralized n8n workflow for processing customer interactions from multiple channels, including webhooks, audio uploads, and inbound email. The system uses n8n to turn raw customer communication into structured, reviewable records. It supports transcription, speaker diarization, summarization, sentiment analysis, and follow-up classification.

The exported n8n workflow can be found in the n8n folder.

Features

• Accepts customer input from:
  • JSON webhooks
  • audio uploads for calls or voicemail
  • inbound email via IMAP
• Transcribes and diarizes audio input
• Normalizes incoming data into a consistent schema
• Generates German summaries using an LLM
• Classifies sentiment as good, neutral, or bad
• Flags follow-up items with open_todo=true
• Flags critical cases with is_critical=true
• Stores processed contacts in the customer_contacts table

Repository Structure

• docker-compose.yml — local stack for all services
• n8n/customer-contact_n8n_workflow.json — exported n8n workflow
• webapp/ — browser UI for sending test text and audio payloads
• audio-processing-api/ — FastAPI service for transcription and diarization

Architecture

Services

• n8n: http://localhost:5678
• webapp: http://localhost:8000
• audio-processing-api: http://localhost:8001
• ollama: http://localhost:11434

Requirements

Before starting, make sure you have:

• Docker
• Docker Compose
• an AssemblyAI API key for speaker diarization

Optional:

• IMAP credentials for inbound email processing
• SMTP or other email credentials for notification steps in n8n

Quickstart

1. Configure environment variables

Copy the example environment file:

cp .env.example .env

Then update the values in .env to match your local setup.

2. Start the services

Start the default stack:

docker compose up --build -d

Stop the stack:

docker compose stop

3. Enable Ollama locally (optional)

If you want to run Ollama locally, start the full profile and pull the model:

docker compose --profile full up --build -d
docker exec -it ollama ollama pull llama3.2:3b

Notes:

• The model download may take some time.
• It requires roughly 10 GB of disk space.

4. Initialize n8n

Open:

http://localhost:5678

Complete the first-time owner setup, then:

1. Open n8n.
2. Import n8n/customer-contact_n8n_workflow.json.
3. Run the Manual Trigger once to create the customer_contacts data table.
4. Configure the credentials referenced by the workflow.
5. Activate the workflow.

Ollama note

If you use the local ollama service, set the base URL of the model credentials to:

http://ollama:11434

If you do not want to use Ollama, replace the LLM node with another supported model provider.

5. Test the setup

Open the test UI:

http://localhost:8000

The UI allows you to:

• send text messages to /webhook-test/... or /webhook/...
• upload audio files to test or production webhook endpoints

Development

Audio Processing API

The audio service exposes:

• POST /api/transcribe Uses faster-whisper with the small model on CPU
• POST /api/diarize Uses AssemblyAI and returns speaker-grouped transcript segments

Web App

The web app is a lightweight FastAPI application that serves a single HTML page for manual testing.

It includes:

• an audio upload form with a phone number field
• a message form with name, email, phone, subject, and message
• a sample-message selector for quickly sending predefined customer requests

Known Limitations

• The demo setup does not include authentication for the UI, webhooks, or n8n.
• The workflow is currently built for German contacts and summaries.
• The bundled llama3.2:3b model may be too weak for reliable structured output.
• Diarization depends on AssemblyAI and requires outbound network access.
• Critical-case detection currently relies on simple rules rather than a more robust decision model.

Testing

There is currently no automated test suite.

For now, validation is manual:

1. Start the stack with Docker Compose.
2. Open the web UI at http://localhost:8000.
3. Submit sample text messages and audio files to the test webhooks.
4. Confirm that contacts are processed and written to the customer_contacts table in n8n.