MiRAG — Multi-input Retrieval-Augmented Generation

MiRAG is an interactive, multi-modal application built with Streamlit that leverages Retrieval-Augmented Generation (RAG) to perform question-answering and summarization across various content types:

• 🌐 Web pages
• 📄 PDF documents
• 📺 YouTube videos
• 📝 Custom user input

Built on LangChain, Gemini (Google Generative AI), and FAISS, MiRAG enables users to query unstructured content intelligently and intuitively.

---

🚀 Features

🔹 Web QA (RAG from URLs)

• Extract and embed content from any public URL (JS and non-JS).
• Perform context-aware question answering and summarization.
• Retain memory across conversation turns.

🔹 PDF QA

• Upload any PDF and perform:
  • Contextual Q&A
  • Full-document summarization
  • Chat history export as PDF

🔹 YouTube Video QA

• Input any YouTube video URL to fetch its transcript.
• Ask questions and generate a summary.
• Ideal for educational content, lectures, and long-form videos.

🔹 Custom Text QA

• Use default chatbot mode or paste your own text block.
• Build a temporary vectorstore and perform RAG on your content.
• Memory support with chat history download.

---

🛠️ Tech Stack

• Python 3.10+
• Streamlit – User Interface
• LangChain – Chain and embedding orchestration
• Google Generative AI (Gemini) – LLM & embeddings
• FAISS – Vectorstore for semantic retrieval
• YouTube Transcript API – Transcript extraction
• FPDF – PDF generation for exporting chats

---

📦 Installation

1. Clone the repository:

   git clone https://github.com/iamtgiri/MiRAG.git
   cd MiRAG

2. Create a virtual environment:

   python -m venv .venv
   source .venv/bin/activate   # On Windows: .venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Set environment variable:

   export GOOGLE_API_KEY=your_api_key_here 

5. Run the app:

   streamlit run app.py

---

📁 Project Structure

MiRAG/
├── app.py                      # Main Streamlit app
├── pdf_utils.py                # PDF loading, splitting & summarization
├── process_youtube.py          # YouTube video processing & transcript extraction
├── rag_utils.py                # Utility functions & chain builders
├── requirements.txt
└── README.md

---

📸 Screenshots

A preview of the MiRAG application in action across different modules:

🏠 Home Interface

Module selection screen and branding

---

📝 Custom Text QA

Normal Q&A without any context Paste custom text, ask questions, and get answers using RAG with memory

---

🌐 Web QA

Enter a URL, extract content, and perform context-aware Q&A

---

📄 PDF QA

Upload a PDF, ask questions, and download the chat history as a PDF:

---

📺 YouTube QA

Enter a YouTube URL, analyze the transcript, and chat with context: Summarize the video and export the chat:

---

📤 Download Chat History

Export your full conversation as a downloadable PDF

---

🧠 Credits

• Built with LangChain
• Powered by Google Gemini
• PDF export via FPDF
• Transcripts via YouTube Transcript API

---

📄 License

MIT License © 2025 Tanmoy Giri
See LICENCE for details.