https://github.com/mj-deving/n8n-meeting-intelligence

Meeting recording → transcript → protocol + action items in 19s. n8n workflow with Whisper speech-to-text and Claude analysis.
https://github.com/mj-deving/n8n-meeting-intelligence

ai-agents code-mode n8n typescript workflow-automation

Last synced: 3 months ago
JSON representation

Meeting recording → transcript → protocol + action items in 19s. n8n workflow with Whisper speech-to-text and Claude analysis.

Host: GitHub
URL: https://github.com/mj-deving/n8n-meeting-intelligence
Owner: mj-deving
License: mit
Created: 2026-04-13T14:07:16.000Z (3 months ago)
Default Branch: main
Last Pushed: 2026-04-22T06:37:56.000Z (3 months ago)
Last Synced: 2026-04-22T08:32:40.729Z (3 months ago)
Topics: ai-agents, code-mode, n8n, typescript, workflow-automation
Language: TypeScript
Size: 59.6 KB
Stars: 0
Watchers: 0
Forks: 0
Open Issues: 0
Metadata Files:
- Readme: README.md
- Funding: .github/FUNDING.yml
- License: LICENSE
- Agents: AGENTS.md

Awesome Lists containing this project

README

          # Meeting Intelligence Pipeline

![n8n](https://img.shields.io/badge/-n8n-EA4B71?style=flat-square) ![TypeScript](https://img.shields.io/badge/-TypeScript-3178C6?style=flat-square) ![Whisper](https://img.shields.io/badge/-Whisper-74AA9C?style=flat-square) ![Claude](https://img.shields.io/badge/-Claude-D4A574?style=flat-square)

**Meeting audio or text transcript → AI analysis → structured protocol with action items, decisions, and follow-ups.** 14-node n8n workflow powered by Claude Sonnet via OpenRouter. Processing in ~19 seconds, outputs to Google Sheets, Gmail, and Slack.

## Table of Contents

- [Architecture](#architecture) — 14-node workflow with 2 trigger variants

- [How It Works](#how-it-works) — pipeline steps in detail

- [Quick Start](#quick-start) — setup in 5 steps

- [Example Output](#example-output) — real Slack message from test run

- [Test Scenarios](#test-scenarios) — 3 verified scenarios

- [LLM Output Schema](#llm-output-schema) — 8 extracted fields

- [Google Sheets Schema](#google-sheets-schema) — 12 CRM columns

- [Tech Stack](#tech-stack) — versions and dependencies

- [Credentials](#credentials) — required n8n credentials

- [Whisper Options](#whisper-options) — API vs. local (GDPR)

- [Project Structure](#project-structure)

## Architecture

```mermaid

flowchart LR

    TextWH["Text Webhook\nPOST /webhook/meeting-text"] --> Analyze

    AudioWH["Audio Webhook\nPOST /webhook/meeting"] --> Whisper["Whisper API\nOpenAI whisper-1"]

    Whisper --> Analyze

    Analyze["Analyze Meeting\n(Claude Sonnet via OpenRouter)"] --> Prepare

    Analyze -.->|AI| Model["Claude Model\nanthropic/claude-sonnet-4"]

    Analyze -.->|Parser| Schema["Meeting Schema\nStructured Output + AutoFix"]

    Prepare["Prepare CRM Data\n(Code Node)"] --> Time["Measure Time"]

    Time --> Sheets["Log to Google Sheets\n(Meeting Intelligence CRM)"]

    Time --> Email["Gmail Send\n(HTML Protocol)"]

    Time --> Slack["Format + Post Slack\n(Action Items)"]

    Time --> Response["Webhook Response\n(JSON Summary)"]

    style TextWH fill:#e1f5fe,color:#0f1117,stroke:none

    style AudioWH fill:#e8f5e9,color:#0f1117,stroke:none

    style Whisper fill:#e8f5e9,color:#0f1117,stroke:none

    style Analyze fill:#fff3e0,color:#0f1117,stroke:none

    style Model fill:#fff3e0,color:#0f1117,stroke:none

    style Schema fill:#fff3e0,color:#0f1117,stroke:none

    style Prepare fill:#f5f5f5,color:#0f1117,stroke:none

    style Time fill:#f5f5f5,color:#0f1117,stroke:none

    style Sheets fill:#e8eaf6,color:#0f1117,stroke:none

    style Email fill:#c8e6c9,color:#0f1117,stroke:none

    style Slack fill:#ffcdd2,color:#0f1117,stroke:none

    style Response fill:#e8eaf6,color:#0f1117,stroke:none

```

| Node | Type | Purpose |

|------|------|---------|

| Text Webhook | webhook | JSON input: title, participants, date, transcript |

| Audio Webhook | webhook | Multipart audio upload |

| Whisper Transcription | httpRequest | OpenAI Whisper API (model: whisper-1, language: de) |

| Analyze Meeting | agent (LangChain) | Claude Sonnet analyzes transcript |

| Claude Model | lmChatOpenAi | OpenRouter → anthropic/claude-sonnet-4 |

| Meeting Schema | outputParserStructured | JSON schema with 8 required fields |

| AutoFix Model | lmChatOpenAi | Gemini Flash for schema auto-fix |

| Prepare CRM Data | code | Formats 12 columns for Google Sheets |

| Measure Processing Time | code | Calculates processing duration |

| Log to Google Sheets | googleSheets | Append row with explicit column mapping |

| Send Protocol Email | gmail | HTML protocol to participants |

| Format Slack Message | code | Slack markdown with priority icons |

| Post Slack Actions | slack | Team channel, @channel on urgent |

| Webhook Response | code | JSON summary as response |

## How It Works

1. **Receive transcript** — Text webhook (JSON) or audio webhook (multipart → Whisper)

2. **AI analysis** — Claude Sonnet extracts: summary, decisions, action items, open questions, follow-ups, key topics, sentiment

3. **Google Sheets** — Protocol as CRM row with 12 columns

4. **Gmail** — HTML-formatted protocol to participants (summary, decisions as bullets, action items as table)

5. **Slack** — Action items with @channel on high-priority items

## Quick Start

```bash

# 1. Install dependencies

npm install

# 2. Connect n8n instance

npx --yes n8nac init

# URL: http://172.31.224.1:5678

# 3. Create Google Sheet

npx --yes n8nac push "workflows/172_31_224_1:5678_marius _j/personal/setup-meeting-sheet.workflow.ts"

npx --yes n8nac workflow activate Cctig8XetXsoKeou

curl http://172.31.224.1:5678/webhook/setup-meeting-sheet

# 4. Deploy main workflow

npx --yes n8nac push "workflows/172_31_224_1:5678_marius _j/personal/meeting-intelligence.workflow.ts"

npx --yes n8nac workflow activate k2VzgzfxKOtosxzn

# 5. Test

curl -X POST http://172.31.224.1:5678/webhook/meeting-text \

  -H "Content-Type: application/json" \

  -d @workflows/pipelines/meeting-intelligence/test.json

```

## Example Output

**Slack message (Sprint Planning Q2):**

```

@channel *📋 Meeting Protocol: Sprint Planning Q2*

_2026-04-14 | Participants: Marius, Lisa, Thomas_

*Summary:* Sprint planning for Q2 focused on the Meeting Intelligence Pipeline

for the conference demo on April 28. Decision to use Claude Sonnet via OpenRouter.

*Action Items:*

  🔴 *Thomas*: Complete API documentation (by this week)

  🔴 *Marius*: Share OpenRouter credentials with Thomas (by today)

  🔴 *Thomas*: Take over Whisper integration (by Friday)

  🔴 *Lisa*: Prepare Google Sheets template (by Wednesday)

  🟡 *Marius*: Implement Slack integration (by Thursday)

  ⚪ *Thomas*: Document local Whisper option

*Decisions:*

  ✅ Prioritize Meeting Intelligence Pipeline as sprint goal #1

  ✅ Use Claude Sonnet via OpenRouter instead of GPT-4

```

## Verified Test Results

| Metric | Value |

|--------|-------|

| Processing time | 19.4 seconds |

| Transcript length | 292 words |

| Action items extracted | 6 |

| Decisions detected | 2 |

| Sentiment | positive |

| All 12 nodes | OK |

| Execution ID | #158 |

| Date | 2026-04-13 |

## Test Scenarios

3 test payloads in `workflows/pipelines/meeting-intelligence/test.json`:

| # | Scenario | Participants | Transcript | Expected |

|---|----------|-------------|------------|----------|

| 1 | Sprint Planning Q2 | Marius, Lisa, Thomas | 292 words | 3+ action items, 2+ decisions, positive |

| 2 | Customer Call (Autohaus Muller) | Marius, Thomas Muller | 266 words | 2+ action items, 1+ follow-up, positive |

| 3 | Incident Review: API Outage | Marius, DevOps Team | 281 words | 4+ action items, 1+ decision, negative |

## LLM Output Schema

```json

{

  "summary": "3-5 sentence summary",

  "decisions": [{"decision": "...", "context": "..."}],

  "action_items": [{"owner": "...", "task": "...", "deadline": "...|null", "priority": "high|medium|low"}],

  "open_questions": ["..."],

  "follow_ups": [{"topic": "...", "when": "...", "participants": ["..."]}],

  "key_topics": ["..."],

  "sentiment": "positive|neutral|negative",

  "duration_estimate_min": 30

}

```

## Google Sheets Schema

| Column | Type | Description |

|--------|------|-------------|

| Timestamp | DateTime | Processing timestamp |

| Meeting_Title | Text | Meeting title |

| Date | Date | Meeting date |

| Participants | Text | Participants (comma-separated) |

| Summary | Text | AI-generated summary |

| Decisions | Text | Decisions (numbered) |

| Action_Items | Text | Action items with owner + deadline |

| Open_Questions | Text | Open questions (numbered) |

| Follow_Ups | Text | Planned follow-ups |

| Sentiment | Text | positive / neutral / negative |

| Transcript_Length | Number | Words in transcript |

| Processing_Time_Sec | Number | Processing duration in seconds |

## Tech Stack

| Component | Version / Detail |

|-----------|-----------------|

| n8n | 2.x (Windows, WSL access via vEthernet bridge) |

| n8nac | 1.5.0 (code-first workflow development) |

| LLM (Analysis) | anthropic/claude-sonnet-4 via OpenRouter |

| LLM (AutoFix) | google/gemini-2.0-flash-001 via OpenRouter |

| Transcription | OpenAI Whisper API (whisper-1, language: de) |

| Outputs | Google Sheets + Gmail + Slack |

| Language | TypeScript (n8nac .workflow.ts) |

## Credentials

| Credential | Type | Purpose |

|---|---|---|

| OpenRouter | openAiApi | Claude Sonnet + Gemini Flash |

| Google Sheets | googleSheetsOAuth2Api | Meeting protocol CRM |

| Gmail | gmailOAuth2 | Protocol delivery |

| Slack Bot | slackApi | Action items channel |

| OpenAI (optional) | httpHeaderAuth | Whisper API (audio path only) |

## Whisper Options

| Option | Use Case | Details |

|--------|----------|---------|

| **OpenAI API** (default) | Demo / fast | `POST api.openai.com/v1/audio/transcriptions`, model: whisper-1 |

| **Local Server** (GDPR) | Production | `docker run -d -p 9000:9000 onerahmet/openai-whisper-asr-webservice` |

Local Whisper Server Setup

```bash

# faster-whisper as Docker service

docker run -d -p 9000:9000 \

  onerahmet/openai-whisper-asr-webservice:latest

# Adjust in workflow:

# Whisper Transcription Node → change URL to:

# http://localhost:9000/asr?language=de&output=json

```

## Workflows

| Workflow | ID | Status | Nodes |

|---|---|---|---|

| Meeting Intelligence Pipeline | `k2VzgzfxKOtosxzn` | Active | 14 |

| Setup Meeting Intelligence Sheet | `Cctig8XetXsoKeou` | Active | 2 |

## Project Structure

```

n8n-meeting-intelligence/

├── workflows/

│   ├── pipelines/meeting-intelligence/

│   │   ├── README.md              # Detailed workflow documentation

│   │   ├── workflow/

│   │   │   ├── workflow.ts        # n8nac TypeScript source (14 nodes)

│   │   │   └── workflow.json      # n8n JSON export

│   │   └── test.json              # 3 test scenarios

│   └── 172_31_224_1.../personal/

│       ├── meeting-intelligence.workflow.ts  # n8nac sync copy

│       └── setup-meeting-sheet.workflow.ts   # Sheet setup helper

├── scripts/

│   ├── new-workflow.sh            # Scaffold new workflows

│   └── check-secrets.sh           # Pre-commit secret detection

├── docs/                          # GitHub Pages + ADRs

├── CLAUDE.md                      # AI agent instructions

├── AGENTS.md                      # n8nac protocol

└── README.md

```

## License

MIT

ecosyste.ms

Data

Tools

Indexes

Applications

Experiments

Awesome

https://github.com/mj-deving/n8n-meeting-intelligence

Awesome Lists containing this project

README