🔍 Advanced RAG Chatbot

A powerful Retrieval-Augmented Generation (RAG) system for processing multiple PDFs and delivering accurate, context-aware responses to user queries. Built in Python, it features GPU-accelerated embeddings, a semantic search engine, and the Mistral model via Ollama, wrapped in an intuitive Streamlit interface.

---

🚀 Features

• 📄 Multi-PDF Support – Upload and process several PDFs at once.
• ⚡ GPU-Accelerated Embeddings – Leverage NVIDIA GPUs for fast processing.
• 🔎 Semantic Search – Uses FAISS for efficient context retrieval with source tracking.
• 🌐 Streamlit Interface – Clean and interactive web UI for document upload, chat, and performance.
• 🧩 Highly Configurable – Adjust chunk size, overlap, batch size, and worker threads.
• 📊 Real-time Metrics – Monitor chunking speed, memory usage, and system stats.
• 💬 Chat History Export – Save and download all chat sessions in JSON format.

---

📁 Repository Info

Field	Details
Repo Name	Advanced-RAG-Chatbot
License	MIT
Language	Python
Model	Mistral via Ollama
Embeddings	intfloat/e5-small (HuggingFace)
Interface	Streamlit
Vector Store	FAISS
PDF Processor	PyMuPDF
Status	🚧 Actively maintained

---

🖥️ Requirements

✅ Hardware

Type	Minimum	Recommended
CPU	4-core	8-core
RAM	8 GB	16 GB
Storage	10 GB	20 GB
GPU (Optional)	❌	✅ NVIDIA (GTX 1060 or higher)

✅ Software

• Python 3.8+
• Git
• CUDA Toolkit 11.2+ (for GPU acceleration)
• Ollama (≥ 0.1.0)

---

⚙️ Installation

1️⃣ Install Python

# Download Python from https://python.org
python --version  # ✅ Ensure it's ≥ 3.8

2️⃣ Install Git

# Download from https://git-scm.com
git --version

3️⃣ Clone Repository

git clone https://github.com/milind899/Advanced-RAG-Chatbot.git
cd Advanced-RAG-Chatbot

4️⃣ Create Virtual Environment

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

5️⃣ Install Dependencies

pip install -r requirements.txt

📦 Key Libraries

• streamlit – Frontend UI
• langchain – RAG orchestration
• torch – Model + CUDA acceleration
• transformers – Embedding generation
• faiss-cpu – Vector similarity search
• PyMuPDF – PDF parser

---

🤖 Ollama Setup

Install Ollama

curl -fsSL https://ollama.ai/install.sh | sh
ollama pull mistral
ollama serve

Verify Setup

curl http://localhost:11434/api/tags

Enable GPU (Optional)

# Verify CUDA
nvidia-smi

# Enable GPU
unset OLLAMA_NO_CUDA

# Test with PyTorch
python -c "import torch; print(torch.cuda.is_available())"

---

🚀 Run the Application

streamlit run app.py

🔗 Open your browser at: http://localhost:8501

---

💡 How to Use

📂 Uploading PDFs

1. Open the app in your browser.

2. Drag-and-drop or browse to upload PDFs.

3. Adjust settings from the sidebar:

   • Chunk Size (default: 500)
   • Overlap (default: 50)
   • Workers (CPU count default)
   • Batch Size (default: 100)

4. Click Process Documents.

❓ Asking Questions

• Go to the Chat with Your Documents section.
• Type a question and hit Send.
• View responses with source links.
• Review chat history at the bottom.
• Export your chats via the Export JSON button.

📈 Monitoring & Stats

• View chunks/sec, total time in Performance Metrics.
• Sidebar shows GPU & system usage.
• Troubleshoot with real-time logs.

---

🗂️ Project Structure

Advanced-RAG-Chatbot/
├── app.py              # Streamlit UI
├── rag_backend.py      # Core RAG logic
├── requirements.txt    # Dependencies
├── LICENSE             # MIT License
├── faiss_index/        # Generated vector store
├── embedding_cache/    # Cached embeddings

---

⚙️ Configuration Options (in rag_backend.py)

Parameter	Description	Default
chunk_size	Text chunk size	500
chunk_overlap	Overlap between chunks	50
max_workers	Concurrent threads	CPU count
batch_size	Documents per embedding batch	100
embedding_model	HuggingFace model	intfloat/e5-small
cache_embeddings	Use embedding cache	True

---

🧯 Troubleshooting

Problem	Fix
❌ Ollama not responding	Ensure ollama serve is running and model is pulled.
❌ CUDA not available	Install CUDA Toolkit, check nvidia-smi, enable with unset.
💥 Memory crash	Reduce batch_size or chunk_size, clear cache folders.
🐢 Slow performance	Enable GPU, increase max_workers, tune chunking strategy.

---

📬 Support

• 🔧 Use the Help & Tips section in the app.
• 🐛 File issues or suggestions on GitHub Issues
• 📄 Refer to rag_backend.py for logic and model details.

---

📜 License

This project is licensed under the MIT License. See LICENSE for more information.