Summarize Zoom meetings with GPT-4o, Whisper, Supabase RAG and email reports

Description

AI Meeting Intelligence System

> Zoom + n8n + GPT-4o + Supabase RAG

This n8n workflow automates Zoom meeting intelligence by capturing recordings via webhook, transcribing audio using OpenAI Whisper, analyzing it with GPT-4o and storing structured insights in Supabase vector database. It detects summaries, decisions, action items and contradictions, then sends an email report to stakeholders.

Quick Implementation Steps

Import workflow json in your n8n account
Connect Zoom webhook for recording events
Configure OpenAI API (Whisper + GPT-4o)
Set up Supabase meeting_memories table with vector support
Add embeddings credentials in n8n
Activate workflow and test with a Zoom recording

What It Does

This workflow automatically processes Zoom meeting recordings and converts them into structured intelligence. It extracts transcript text, analyzes it using GPT-4o and generates structured insights like decisions, summaries and action items.

It also compares new meetings with historical data using Supabase vector search to detect contradictions or repeated decisions.

Finally, it stores everything in a semantic memory database and sends a clean email report to stakeholders.

Who It's For

Product Managers
Engineering Teams
Startup Founders
Project Managers
AI Automation Engineers
Remote teams using Zoom

Requirements to Use This Workflow

n8n account (cloud or self-hosted)
Zoom account with webhook access
OpenAI API key (Whisper + GPT-4o)
Supabase project with vector enabled
Gmail OAuth credentials
Public webhook URL

How It Works & Setup Guide

1. Zoom Ingestion Layer

Zoom webhook captures meeting recording events and sends metadata like recording URL and meeting title to n8n.

2. Data Extraction Layer

The workflow extracts recording URL and metadata using a set node.

3. Audio Processing Layer

Audio is downloaded from Zoom cloud and sent to OpenAI Whisper for transcription into text.

4. Transcript Normalization Layer

The raw transcript is cleaned and formatted into a structured format for AI processing.

5. AI Intelligence Layer (GPT-4o)

GPT-4o analyzes the transcript and extracts:

Summary
Decision
Action items
Contradiction report

Structured output parser ensures valid JSON format.

6. Deduplication Layer

A hash is generated from the transcript and checked in Supabase:

If duplicate exists → stop workflow
If not → continue processing

7. RAG Memory Layer

Embeddings are created and stored in Supabase vector DB. Past meetings are retrieved to compare decisions and detect contradictions.

8. Notification Layer

Final structured output is stored and emailed to stakeholders with insights.

How To Customize Nodes

Modify GPT prompt for domain-specific intelligence
Extend metadata with project/client IDs
Replace Gmail with Slack or Teams integration
Add Jira task creation for action items
Customize Supabase schema for multi-team support

Add-ons

Slack real-time meeting alerts
Jira auto task creation
Dashboard using Supabase + Next.js
Multi-language transcription support
Speaker identification using Whisper timestamps

Use Case Examples

Automated meeting minutes generation
Engineering decision tracking system
Product requirement documentation automation
Client meeting summaries
Compliance audit trail generation

More use cases can be built depending on customization.

Troubleshooting Guide

| Issue | Possible Cause | Solution |
|------|---------------|----------|
| No transcript generated | Invalid Zoom recording URL | Check webhook payload |
| Duplicate not detected | Incorrect Supabase query path | Fix metadata filter |
| Empty AI response | Weak prompt structure | Improve prompt clarity |
| Email not sent | Gmail auth expired | Reconnect Gmail OAuth |
| Supabase returns empty | Wrong filter syntax | Validate JSON column query |

Need Help

If you need help customizing or scaling this workflow into production-grade automation, you can get expert assistance from n8n developers at WeblineIndia. They specialize in building AI-powered automation systems using n8n, OpenAI and vector databases.