Skip to content

README.md

Latest commit

 

History

History
103 lines (88 loc) · 4.43 KB

README.md

File metadata and controls

103 lines (88 loc) · 4.43 KB
Raw

VidSynth AI

VidSynth AI is a YouTube content synthesizer that turns any public video into structured knowledge. It can summarize long-form content into bullet-point notes, extract the most important topics, or let you chat with the video through a Retrieval-Augmented Generation (RAG) workflow powered by Gemini models.

Features

  • Transcript ingestion Fetches captions for a YouTube video using youtube-transcript-api, handling multiple language codes.
  • On-demand translation Converts non-English transcripts to English with the Gemini gemini-2.5-flash-lite chat model.
  • Summaries and notes Produces human-friendly highlights and study notes tailored to the video content.
  • Topic extraction Distills the top five topics discussed in the source video.
  • Conversational RAG chatbot Splits transcripts into chunks, stores them in an in-memory Chroma vector store, and answers questions about the video context.
  • Multiple runtime options Includes a Streamlit interface (app.py) and a production-ready Flask backend (flask_app.py) with static assets (index.html, style.css, script.js).

Project Structure

project-root/
├── app.py                  # Streamlit front-end
├── flask_app.py            # Flask API + static site
├── supporting_functions.py # Transcript, translation, RAG utilities
├── index.html              # Front-end served by Flask
├── style.css
├── script.js
├── requirements.txt
├── runtime.txt             # Python version for deployment
├── Procfile                # Process definition for Render/Heroku-like platforms
├── render.yaml             # Render.com deployment blueprint
├── simple_test.py          # Local sanity test for vector store creation
└── ...

Prerequisites

  • Python 3.11.x (see runtime.txt)
  • A Google Generative AI API key with access to Gemini models
  • pip for dependency management

Setup

  1. Clone and enter the project

    git clone <your-fork-url>
cd project-3-youtube-rag

  2. Create a virtual environment (recommended)

    python -m venv .venv
.venv\Scripts\activate  # Windows

  3. Install dependencies

    pip install -r requirements.txt

  4. Configure environment variables Create a .env file in the project root with your API key:

    GOOGLE_API_KEY=your-google-generative-ai-key

    Important: Do not commit the .env file. The included .env is for local reference only—replace the placeholder key with your own credentials.

Running the Application

Option 1: Streamlit interface

  1. Start the Streamlit app: 
    streamlit run app.py
  2. Open the provided local URL in your browser.
  3. Paste a YouTube link, choose the language code, and select either Chat with Video or Notes For You.

Option 2: Flask API + static front-end

  1. Ensure the .env file is present.
  2. Launch the Flask server: 
    python flask_app.py
  3. Navigate to http://localhost:5000 for the full web experience, or interact with the REST endpoints directly (e.g., POST /api/transcript, POST /api/chat).

Testing

Run the lightweight vector store sanity check with:

python simple_test.py

This script validates that chunking, embedding creation, and similarity search succeed with the current dependencies.

Deployment (Render.com example)

  • render.yaml defines a starter Render service that installs requirements and starts flask_app.py.
  • Procfile contains the same start command for Heroku-style platforms.
  • Add GOOGLE_API_KEY as a Render environment variable (marked sync: false to populate it manually in the dashboard).

Troubleshooting

  • Invalid YouTube URL: Ensure the link is public and contains a valid video ID. The helper extract_video_id() in supporting_functions.py expects standard YouTube URL formats.
  • Transcript fetch failures: Some videos disable transcripts or restrict them by region/language. Try another language code or video.
  • Gemini API errors: Verify your API key, project quota, and network connectivity.
  • Chat responses missing: Confirm the vector store was created. In Streamlit, the chatbot becomes available after the transcript is processed and chunks are indexed.

License

Specify your preferred license here before publishing (e.g., MIT, Apache 2.0). Remember to add a dedicated LICENSE file if needed.