🎬 YouTube KPT Synthesizer AI Agent with Multi-LLM

This repository contains a production-ready n8n workflow that ingests YouTube video content, extracts metadata and transcripts, cleans and structures the data, and synthesizes insights using multiple Large Language Models (LLMs) such as OpenAI GPT-4 and DeepSeek. This module implements a robust Knowledge, Process, Technology (KPT) summarizer pipeline for actionable insights.

---

🧠 Project Overview

The YouTube KPT Synthesizer AI Agent automates end-to-end video insight generation with the following architecture:

1. Ingest YouTube Data

   • Fetches video metadata (videoId, channelName, publishDate, fullUrl) via YouTube Data API.
   • Retrieves full transcripts using the Apify YouTube Scraper API.

2. Clean and Process

   • Cleans transcript data, removing noise, and consolidates it into a structured fullTranscript.
   • Merges metadata and transcript into a unified, consistent data structure.

3. Multi-LLM Synthesis

   • Uses OpenAI GPT-4 to generate structured markdown summaries including:
     • SEO Title
     • One-Sentence Summary
     • Key Takeaways (Knowledge, Process, Technology insights)
     • Concluding Thought
   • Integrates DeepSeek as the final voting and synthesis agent, combining outputs from multiple models into a single actionable KPT summary.

4. Output

   • Posts the final synthesized markdown summary to Slack for immediate team review and archival.
   • Saves the output into connected knowledge bases (e.g. Notion) for long-term strategic tracking.

---

🖼️ Workflow Architecture

YouTube_KPT_Synthesizer_Workflow/ ├── Trigger: Manual or Scheduler ├── Set Node: Input YouTube URL ├── Code Node: Extract Video ID ├── HTTP Request: Fetch YouTube Metadata │ └── Code Node: Extract Metadata (channelName, publishDate, videoId, fullUrl) ├── HTTP Request: Request YouTube Transcript (Apify) │ └── Code Node: Clean Transcript ├── Merge Node: Combine Metadata + Transcript ├── Multi-LLM Synthesis: │ ├── OpenAI GPT-4 Node: Generate KPT Structured Summary │ └── DeepSeek Agent Node: Final Synthesis & Voting (multi-LLM ensemble) └── Slack Node: Post Final Markdown Summary

---

🔥 Key Features

• Multi-LLM Ensemble Reasoning: Combines outputs from GPT-4, Claude Mini, DeepSeek, and others for robust summarization.
• Automated Knowledge Extraction: Converts unstructured transcripts into structured KPT summaries.
• Slack Integration: Delivers insights directly to your Slack channels for rapid decision-making.
• Production-Ready Modular Design: Integrates seamlessly within the broader Synth AI Synthesis Agent framework.

---

🚀 Getting Started

📝 Prerequisites

• n8n (Cloud or self-hosted)
• API keys for:
  • YouTube Data API
  • Apify
  • OpenAI or your preferred OpenRouter models
  • Slack webhook

---

💻 Installation

1. Clone this repository:

git clone https://github.com/your-username/youtube-kpt-synthesizer-ai.git
cd youtube-kpt-synthesizer-ai

2. Import the YouTube_Agent_Clean.json workflow into your n8n instance.

3. Configure your credentials for all API calls within n8n.

4. Execute manually or set a scheduler for automated runs.

---

🤝 Contributing

Contributions are welcome to improve modularity, model integration, or output formatting. Please open an issue or submit a pull request.

---

📄 License

MIT

---

© 2025 | Developed as part of the Synth AI Synthesis Agent system – demonstrating robust, multi-LLM agent orchestration and KPT summarization pipelines.