PDF Brainbox

PDF Brainbox is a Streamlit web application that uses Retrieval-Augmented Generation (RAG) to help users interact intelligently with PDF content. Upload educational slides, books, or notes in PDF format and explore features like Q&A, quizzes, summaries, flashcards, and more — all powered by modern AI.

Live Demo: https://pdf-brainbox.streamlit.app

---

Features

• PDF reading
• AI-powered Q&A chatbot
• Automatic topic summaries
• Auto-generated multiple-choice quizzes
• Interactive flashcards
• Study guide and structured notes

---

Tech Stack

Component	Technology
Frontend	Streamlit
Backend	Python
PDF Parsing	PyPDF2
Embeddings	FAISS + Mistral AI
Data Handling	NumPy

---

How It Works

1. PDF Upload: The app extracts and processes section-based content using PyPDF2.
2. Embedding & Indexing: It generates vector embeddings and indexes them using FAISS.
3. RAG Querying: Questions are answered using relevant chunks retrieved from the vector store and passed to Mistral AI.
4. Content Generation: Summaries, quizzes, and flashcards are created using context-aware prompts.
5. Streamlit Interface: The app displays all outputs through an interactive and responsive interface.

---

Evaluation Metrics & Performance

The RAG pipeline has been quantitatively benchmarked using an automated LLM-as-a-Judge framework evaluated against educational documents (sample.pdf), utilizing FAISS vector retrieval and Mistral AI models (mistral-embed & mistral-small-latest):

Metric	Score	Explanation
Faithfulness (Anti-Hallucination)	1.00 / 1.0 (100%)	Generated responses are strictly grounded in retrieved PDF context without hallucinated or external facts.
Answer Relevance	1.00 / 1.0 (100%)	Generated answers directly and concisely address user queries without unnecessary filler or deviation.
Context Relevance (Precision)	0.95 / 1.0 (95%)	Vector similarity search consistently retrieves document chunks containing the necessary and factual evidence to answer queries.
Avg. Retrieval Latency	~1.50s / query	Time taken for query embedding generation and FAISS vector index retrieval ($k=2$).
Avg. Generation Latency	~0.96s / answer	Time taken for context-augmented response completion by Mistral AI.

---

Installation

To run the app locally:

1. Clone the repository

git clone https://github.com/Uhashini/PDF-Brainbox.git
cd PDF-Brainbox

2. Create a virtual environment (optional)

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

3. Install dependencies

pip install -r requirements.txt

4. Run the Streamlit app

streamlit run app.py