🎥 YouTube TechTalk-to-Article

Transform YouTube technical presentations into premium, structured Markdown articles using Google's Gemini models. This tool uses a multi-agent pipeline to ensure technical accuracy, speaker attribution, and professional narrative flow.

🏗️ Architecture

graph TD;
    A[YouTube URL] -->|yt-dlp| B(.srt Transcript)
    B --> C(Pre-processing)
    C -->|Cleaned Text| D[Agent 1: Diarizer]
    D -->|JSON Speaker Tags| E[Agent 2: Cleaner]
    E -->|Polished Content| F[Agent 3: Architect]
    F -->|Structural Blueprint| G[Agent 4: Writer]
    E -.->|Full Context| G
    G --> H(Final Markdown Article)

Loading

🛠️ Prerequisites

• Node.js (v18+)
• yt-dlp (Available in PATH)
• FFmpeg (Required for subtitle conversion)
• Google Gemini API Key (Get one here)

🚀 Quick Start

1. Clone & Install:

   git clone https://github.com/idshdx/Youtube2Article.git
   cd Youtube2Article
   npm install

2. Configure:

   Create a .env file:

   GEMINI_API_KEY=your_api_key_here

3. Run:

   npm start <YouTube URL> [output_name]

Example

npm start https://youtu.be/j3AUC0x_ju8 intro-mixnets

📦 Pipeline Output

The process generates resources in the /dist folder:

• *_cleaned.txt: The polished, diarized transcript.
• *.md: The final high-quality technical article.

🧠 How it Works

1. Extraction: Uses yt-dlp to fetch the auto-generated English transcript as an .srt file.
2. Pre-processing: Cleans the raw SRT formatting and shapes the text for the pipeline.
3. Diarization: The Diarizer Agent identifies speaker changes and labels segments (Host, Speaker, Audience).
4. Refinement: The Cleaner Agent receives the JSON response, removes verbal fillers, and performs rolling deduplication of auto-caption errors.
5. Architecting: The Architect Agent analyzes the polished text to create a structural "Blueprint" with section titles and detailed summaries.
6. Synthesis: The Writer Agent uses the Architect's blueprint alongside the full Cleaner output to synthesize a cohesive, authoritative technical article.

---

Next: Video frame change recognition and OCR patterns for automatic graphic asset embedding.