📚 RAG Chatbot 🔍 🤖 — Powered by LLaMA 3

Welcome to Deepan's RAG Chatbot, a lightweight, efficient Retrieval-Augmented Generation (RAG) chatbot that supports PDF, text, and image (OCR: Optical Character Recognition) uploads. This app uses the power of LLaMA 3 hosted via Groq API, FAISS for vector search, and Sentence Transformers for document embeddings.

🔗 Hosted on GitHub: Deeps72-ux/RAG-model

---

🚀 Features

• 📁 Upload and process multiple file types: .txt, .pdf, .jpg, .jpeg
• 🔍 OCR support using Tesseract for image files
• 🧠 Embed documents using all-MiniLM-L6-v2 from Sentence Transformers
• 📚 FAISS vector store for fast semantic search
• 💬 Interactive Streamlit chat interface with chat history memory
• 🤖 Contextual responses generated via Groq-hosted LLaMA 3 (8B)

---

🧰 Tech Stack

Component	Library/Tool
User Interface	Streamlit
Embeddings	Sentence Transformers (all-MiniLM-L6-v2)
Vector database	FAISS
OCR	PyTesseract
PDF Reading	PyPDF2
LLM	Groq API (LLaMA 3)
Image Handling	Pillow (PIL)
API Secrets	dotenv (.env file for config)

• For text processing: nltk library was used

---

📁 File Structure

RAG-model/
├── final_RAG_model.py     # 🎯 Main Streamlit app to run the RAG pipeline
├── requirements.txt       # 📦 List of required Python packages
├── .env                   # 🔐 Environment file containing the Groq API key (excluded from Git)
├── README.md              # 📘 Project overview and usage instructions
└── Construction/          # 🏗️ Modular components used to build the final RAG model

---

graph TD
    A[React Frontend] -->|REST API| B[FastAPI Backend]
    
    subgraph Frontend
        A --> C[SourcesPanel: Upload/Manage Sources]
        A --> D[ChatWindow: Display Conversation]
        A --> E[ChatInput: Query Input]
        A --> F[MediaRenderer: Render Images/Videos]
    end
    
    subgraph Backend
        B --> G[Ingestion Service]
        B --> H[Query Processor]
        B --> I[Edge Data Service]
        
        G -->|PDF, Text| J[PyMuPDF/pdfplumber]
        G -->|Audio/Video| K[Whisper/MoviePy]
        G -->|URLs| L[Requests/BeautifulSoup]
        G -->|Google Drive| M[Google Drive API]
        G --> N[Sentence Chunking: NLTK/Spacy]
        G --> O[Embedding: Sentence-Transformers/CLIP]
        O --> P[FAISS Vector Store]
        
        H -->|Context| Q[Gemini-1.5 Pro/Gemini-flash2.0]
        H -->|Web Search| R[SerpAPI]
        H -->|Retrieval| P
        
        I -->|API Calls| S[Edge Data API]
        B --> T[SQLite: Metadata Storage]
    end

Loading

graph TD
    A[📥 Input] --> B1[🔄 Extract Tables (pdfplumber)]
    A --> B2[🔄 Extract Text (PyMuPDF)]
    A --> B3[🔄 OCR/Visual Extract (Donut/CLIP)]
    B1 --> C[🔄 Chunk & Embed]
    B2 --> C
    B3 --> C
    C --> D[💾 FAISS / Vector DB (Euc/384-d)]
    D --> E[🔄 Query Embedding]
    E --> F[🔎 Search Relevant Chunks]
    F --> G[🧠 RAG-based LLM Response]
    G --> H[📤 Response/Answer]
end

Loading

🧪 How It Works

1. Document Upload
   Upload .pdf, .txt, or image files (.jpg, .jpeg). The document uploaded by the user gives the context for the LLM (Large Language Model) to process the user queries

2. Text Extraction

   • Text files are parsed directly.
   • PDF files are parsed with the help of PyPDF2 library
   • Images use Tesseract OCR to extract text.

3. Embedding
   The text extracted is broken down into a list of sentences and converted into a vector notation (process called as embedding). The documents are embedded using all-MiniLM-L6-v2 via sentence-transformers and stored in the faiss (Facebook AI Similarity Search) vector database.

4. FAISS Indexing
   Embeddings are indexed using faiss.IndexFlatL2.

5. Retrieval and generation
   For each user query:

   • The top-3 relevant chunks of text are retrieved from the FAISS vector database. They are found using the euclidean distance similarity i.e. the three vectors nearby the query vector in the FAISS database
   • Feed the query to the LLM (Large Language Model) along with the retrieved text as context
   • Display the result of the LLM

 
            ┌──────────────┐
            │ Context Doc  │
            └──────┬───────┘
                   ↓
     ┌────────────────────────────┐ 
     │   Sentence Transformer     │
     └────────────┬───────────────┘
                  ↓
            ┌──────────────┐
            │   FAISS DB   │◄─────------------ User Query
            └─────┬────────┘                     │
                  ↓ Top-k Relevant Context       │
        ┌──────────────────────────┐             │
        │    Prompt Generator      │◄─────-------┘
        │   (FLAN-T5 / LLaMA-3)    │
        └────────────┬─────────────┘
                     ↓
               ✨ Final Answer ✨
 

---

📝 Setup Instructions

1. Clone the Repository

git clone https://github.com/Deeps72-ux/RAG-model.git
cd RAG-model

2. Create and Activate a Virtual Environment (Optional)

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

3. Install Requirements

pip install -r requirements.txt

4. Install Tesseract OCR

Ubuntu/Debian:

sudo apt install tesseract-ocr

macOS:

brew install tesseract

Windows: Download and install from: https://github.com/tesseract-ocr/tesseract/wiki

5. Run the application

streamlit run final_RAG_model.py

🔐 Environment Variables

Create a .env file in the root of your project:

GROQ_API_KEY=your_groq_api_key_here

⚠️ The file .env is ignored in this repository on accounts of privacy

▶️ Running the App

streamlit run app.py

The page opens with:

Deepan's 🙂  RAG Chatbot 🔍 🤖 
📎 Kindly upload text, PDF, or image files to provide a background context

💡 Sample Workflow

Upload one or more .txt, .pdf, or .jpg files.

You will see the text generated format which you can edit. If you generated text is ok, click "📚 Process Files" to embed them using Sentence Transformers.

Ask any question in the chat box.

Get a concise and relevant answer generated with RAG + Groq's LLaMA 3.

Clear Chat: Option to clear the chat history.

🧪 Example Use Cases

Use the RAG model to understand your academic PDFs or scanned notes.

Ask questions from invoices or image-based documents.

Build personal AI assistants for any kind of domain corpus.

🛠️ Troubleshooting

Tesseract not found? Ensure it's installed and added to your system PATH.

FAISS errors on Windows? Prefer using faiss-cpu or run in WSL for full compatibility.

Groq key errors? Double-check the .env key and verify your usage/quota on Groq.

📊 Checking Groq API Usage As of now, Groq does not provide a public dashboard for token usage like OpenAI. However, you can:

Check limits via email/Groq support.

Track approximate usage manually in the app logs.

Stay tuned for Groq dashboard updates: https://console.groq.com

🔮 Future Improvements

LangChain integration Multi-model RAG

Streaming responses

Highlight matched text in source documents

UI themes and dark mode

🙏 Acknowledgements

Groq

Meta's LLaMA 3

Sentence Transformers

Streamlit

Tesseract OCR

📜 License

Unlicensed

🤝 Contributing

Pull requests, feedback, and ideas are most welcome! Just fork the repo, make your changes, and raise a PR.

👋 Author

Deepan

---