RAG System for PDFs and YouTube Videos

This is a multi-page Streamlit application that implements a Retrieval-Augmented Generation (RAG) system using MLX for optimized LLM inference on Apple Silicon. It allows users to analyze PDFs and YouTube videos by extracting and querying relevant content.

---

Features

1. Home Page

• Serves as the main entry point for the application.
• Initializes the shared RAG system.
• Displays system status and navigation options.

2. PDF Analysis Page

• Upload and process PDF documents.
• Query documents with page filtering to refine search results.
• View answers with source references from the document.

3. YouTube Analysis Page

• Process YouTube videos via URL.
• Download and transcribe videos using OpenAI Whisper.
• Extract text from slides in the video using OCR.
• Manually input transcripts for better control over the data.

---

Enhancements & Improvements

This project improves upon a standard RAG system with:

✅ Modular Design – Extracted RAG system into a shared module for reuse.
✅ Common Query Interface – Unified querying for PDFs and YouTube transcripts.
✅ Manual Transcript Management – Users can provide their own transcripts.

---

Installation

1. Clone the repository

   git clone <your-repo-url>
   cd <your-repo-directory>

2. Install dependencies

   pip install streamlit pytube openai opencv-python pytesseract pillow langchain langchain_huggingface langchain_chroma pypdf mlx mlx_lm

3. Download the MLX model

   huggingface-cli download mlx-community/DeepSeek-R1-Distill-Qwen-14B-4bit --local-dir ~/mlx-models

4. Run the application

   streamlit run Home.py