📘 PDF Knowledge Assistant

A chat-based Retrieval-Augmented Generation (RAG) application that allows users to ask questions directly from PDF documents and receive accurate, document-grounded answers with page-level citations and relevance scoring.

This project is designed with a strong focus on trust, transparency, and usability, ensuring that answers are generated strictly from the uploaded document — without hallucinations or external knowledge.

---

🚀 Features

• 📄 Upload any PDF and interact with its content
• 💬 Chat-style conversational interface
• 📌 Page-level citations for each answer
• 🎯 Relevance scoring with visual indicators
• 🛡️ Strictly document-grounded answers (no hallucinations)
• 🎨 Clean, professional, and intuitive UI
• 🔐 Secure handling of API keys (no hardcoding)

---

---

🌐 Live Demo

🔗 Live Application

https://pdf-assistent.streamlit.app/

---

🧠 How It Works

1. PDF Loading
   The uploaded PDF is read page-by-page while preserving page numbers.

2. Text Chunking
   Each page is split into meaningful, sentence-based chunks while keeping metadata such as the page number.

3. Embedding Generation
   Each chunk is converted into a semantic vector using a Sentence Transformer model.

4. Vector Storage
   All embeddings are stored in a ChromaDB vector database along with their metadata.

5. Semantic Retrieval
   When a question is asked, the most relevant chunks are retrieved using vector similarity search.

6. Answer Generation
   A language model generates answers strictly from the retrieved chunks. If the answer is not present in the document, the assistant responds accordingly.

7. Explainability
   The UI displays:

   • Page numbers for source content
   • Relevance scores for each retrieved chunk
   • Optional visibility into the supporting text

---

🧩 Architecture Overview

PDF Document ↓ Page-wise Parsing ↓ Sentence Chunking ↓ Vector Embeddings ↓ ChromaDB Vector Store ↓ Semantic Retrieval ↓ LLM Answer Generation ↓ Chat UI with Citations & Scores

🛠️ Tech Stack

• Frontend / UI: Streamlit
• Vector Database: ChromaDB
• Embeddings: Sentence Transformers (all-MiniLM-L6-v2)
• LLM Inference: Hugging Face Inference API
• PDF Processing: PyMuPDF
• Language: Python

---

🔐 Security & Best Practices

• API keys are never hardcoded
• Secrets are managed using environment variables
• .env files are excluded using .gitignore
• Deployment secrets are stored using platform secret managers
• Clean separation between UI, retrieval, and generation logic

---

📂 Project Structure

PDF-Knowledge-Assistant/
assets/
├── question_demo.png
└── conversation_demo.png
│
├── app.py                # Streamlit application
├── requirements.txt      # Python dependencies
├── README.md             # Project documentation
├── .gitignore            # Ignore secrets & cache files
│
├── rag/
│   ├── loader.py         # PDF loading with page metadata
│   ├── chunker.py        # Sentence-based text chunking
│   ├── vectorstore.py    # Vector storage & retrieval logic
│   └── qa.py             # Answer generation using LLM

🧪 Example Use Cases

• Studying from textbooks and research papers
• Quickly finding information in large PDFs
• Verifying answers with exact page references
• Understanding and summarizing technical documents

---

🎯 Why This Project Matters

Unlike generic AI chatbots, this assistant:

• Does not hallucinate
• Produces verifiable answers
• Clearly shows where each answer comes from
• Emphasizes explainable and trustworthy AI

This project demonstrates a real-world RAG system with production-oriented design, not a prompt-only chatbot or tutorial example.

---

📌 Future Enhancements

• Inline citations within generated answers
• Export chat history as PDF
• Support for multiple PDFs
• Highlighting answer sentences in source text

---

🧠 Author Notes

This project was built with a focus on clean architecture, robustness, and user trust. It reflects production-level thinking and practical application of Retrieval-Augmented Generation systems.

---

⭐ If you find this project useful, consider starring the repository!