RAG PDF Chatbot - 100% Local

This is a Streamlit-based chatbot that allows users to upload PDFs and ask questions about the content. It runs entirely locally, meaning no API costs, no internet requirements, and full privacy.

---

Workflow

Features

• 📂 Upload multiple PDFs
• 🔍 Vector search using FAISS
• 🧠 Embeddings with Hugging Face models
• 🤖 Uses Ollama (Llama 2) for local inference
• ⚡ No API keys required
• 🔒 Runs 100% on your local machine

---

🛠 Installation

1️⃣ Create a Virtual Environment (Optional but recommended)

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

2️⃣ Install Dependencies

Using pip

pip install -r requirements.txt

Using conda (for CPU versions only)

conda install -c conda-forge pytorch transformers sentence-transformers faiss-cpu
pip install -r requirements.txt

---

🏗 Project Structure

📂 pdf-chatbot
├── 📄 app.py               # Main Streamlit app
├── 📂 modules
│   ├── process_pdf.py      # PDF processing (text extraction, chunking)
│   ├── vector_store.py     # FAISS vector database management
│   ├── llm_inference.py    # Running Ollama for answering queries
│   ├── config.py           # Configurations (chunk size, model settings)
│   ├── __init__.py         # Module initialization
├── 📂 faiss_index          # Saved FAISS index (generated after processing PDFs)
├── 📄 requirements.txt      # Dependencies
├── 📄 README.md            # This file

---

▶️ How to Run

streamlit run app.py

Then open http://localhost:8501 in your browser.

---

📝 Usage

1. Upload one or more PDF files from the sidebar.
2. Set the chunk size (default: 300).
3. Click "Process Documents" to generate embeddings.
4. Enter your question in the text box.
5. Get an AI-powered answer based on the PDFs!

---

🏗 Technologies Used

• Streamlit (Web UI)
• PyPDF2 (PDF text extraction)
• LangChain (Text chunking, LLM inference)
• FAISS (Vector search)
• Hugging Face Embeddings (Text representations)
• Ollama (Llama 2) (Local LLM for answering queries)

---

⚠️ Troubleshooting

• No answer or incorrect response?
  • Ensure PDFs contain selectable text (scanned images won’t work)
  • Increase chunk size if responses lack context
• Performance issues?
  • Reduce chunk size for faster processing
  • Use tinyllama instead of llama2 for quicker inference

---

Future Enhancements

• ✅ Add support for PDFs with scanned text (OCR)
• ✅ Implement multi-document summarization
• ✅ Improve LLM response formatting and citations

---

💡 Credits & License

Developed by Swagath Babu. Free for personal and research use.