📝 PDF QnA Helper – Local RAG Bot

🚀 Overview

This is a lightweight, privacy-focused RAG (Retrieval-Augmented Generation) bot designed to help with research papers, coursework, and large PDFs—without relying on external APIs.

Why?

As a bioinformatics graduate, reading dense research papers was frustrating, especially with large, complex PDFs. LLMs can help, but most require cloud processing, raising privacy concerns for copyrighted coursework. This project solves that by keeping everything local while ensuring efficient QnA responses.

💡 Note: This project requires at least 8GB of RAM to run smoothly. Lower memory may lead to performance issues.

🏗️ Reasoning

Component	Model/Tool Used	Why?
LLM	mistral-7b-v3 (via Ollama)	Open-source, fast, and outperforms larger models in efficiency
Embeddings	all-mpnet-base-v2	Lightweight and powerful for sentence embeddings
Vector DB	FAISS	Efficient similarity search
Retrieval	Cosine Similarity Search	Finds the most relevant PDF chunks
Processing	PyMuPDF / pdfplumber	Handles complex PDFs

---

📂 Directory Structure

├──   data/                 # Stores PDFs (e.g., “biochem.pdf”)
├──   src/                  # Main application source code
│   ├── main.py             # Entry point for the app
│   ├── models.py           # Manages LLM and embedding model
│   ├── pdf_processing.py   # Handles PDF text extraction & chunking
│   ├── query_script.py     # Handles user queries (RAG pipeline)
├── .gitignore              # Ignores cache, pycache, and other unnecessary files
├── README.md               # Project documentation
├── requirements.txt        # Dependencies

🔧 Installation & Setup

1️⃣ Clone the repo

git clone hhttps://github.com/A1pha-Z3r0/BioRAG.git
cd BioRAG

2️⃣ Install dependencies

pip install -r requirements.txt

3️⃣ Download models from Ollama

ollama pull mistral

4️⃣ Run the src/main.py script

python src/pdf_processing.py -p data/biochem.pdf

5️⃣Ask Question

Enter your question: _________

💾 System Requirements

To run this project efficiently, your system should have: • ✅ At least 8GB of RAM (recommended for smooth execution) • ✅ A CPU with AVX support (for optimal FAISS performance) • ✅ Ollama installed to run Mistral locally

📌 Features

✅ Fully Local RAG QnA – No cloud dependency ✅ Handles Large PDFs – Efficient text chunking & retrieval ✅ Lightweight Embeddings – Keeps resource usage low ✅ Fast Inference – Uses Mistral 7B for speed and efficiency ✅ Privacy-Focused – No data leaves your machine