MrMrsGaoRAG

A project to get all info from YouTube channel "Mr & Mrs Gao" and build a RAG (Retrieval-Augmented Generation) application.

🌟 Features

• Audio Transcription - Convert YouTube videos to text using Whisper ASR
• Vector Search - FAISS index with M3E embeddings for semantic search
• Multi-LLM Support - Choose between:
  • Local models via Ollama
  • Cloud APIs (GPT-4/DeepSeek)
• Interactive UI - Gradio-based interface with RAG capabilities

🛠️ Installation

# Clone repository
git clone https://github.com/yourusername/MrMrsGaoRAG.git
cd MrMrsGaoRAG

# Install dependencies
pip install -r requirements.txt

# Install Whisper (requires FFmpeg)
brew install ffmpeg  # macOS
# or
sudo apt update && sudo apt install ffmpeg  # Linux

📁 Project Structure

.
├── audios/          # Raw audio files
├── transcribe/      # Whisper JSON transcripts
├── faiss/           # Vector index & metadata
├── query.py         # Main RAG interface
├── transcribe.py    # Audio processing
└── insert_faiss.py  # Index builder

🧠 How It Works

RAG Pipeline

1. Query Processing User input → M3E embedding → FAISS similarity search
2. Context Augmentation Top-K relevant segments + Original query → LLM prompt
3. Response Generation Local/Cloud LLM generates final answer with video sources

Technical Stack

Component	Technology
Transcription	Whisper Medium (Chinese)
Embeddings	M3E-base
Vector Store	FAISS
LLM Interface	LangChain + Ollama
UI Framework	Gradio

⚠️ Notes

• First run will download ~3GB Whisper model
• GPU recommended for faster transcription
• Edit split_text() in insert_faiss.py to adjust chunk sizes
• Add API credentials in Gradio UI when using cloud models