🔎 Queryo Docs

📄 AI-Powered Document Question Answering System using Retrieval-Augmented Generation (RAG)

---

📌 Project Overview

Queryo Docs is an intelligent document-assistant that enables users to "chat" with their PDF documents. By leveraging the power of Retrieval-Augmented Generation (RAG), it allows you to ask natural language questions about any uploaded PDF and receive precise, context-aware answers instantly.

Instead of scanning hundreds of pages manually, Queryo Docs finds the specific sections relevant to your query and uses an AI model to synthesize a clear response.

---

🧐 What is RAG? (The Magic Behind the App)

Retrieval-Augmented Generation (RAG) is a technique used to give LLMs (Large Language Models) access to specific, up-to-date information that wasn't in their original training data.

🧩 How it Works in Simple Terms:

1. Retrieval: When you ask a question, the system searches through your document to find the most relevant paragraphs.
2. Augmentation: These relevant paragraphs are then "attached" to your original question as background context.
3. Generation: The AI model reads both your question AND the context to generate a factual, grounded answer.

This ensures the AI doesn't "hallucinate" (make things up) and instead bases its answers purely on the document you provided.

---

🧠 Key Technologies Explained

To make this happen, Queryo Docs uses several advanced AI concepts:

• Vector Embeddings: These are mathematical representations of text. Think of them as "coordinates" in a space where similar meanings are close together.
• Semantic Search: Unlike keyword search (which looks for exact words), semantic search looks for meaning. If you search for "pay," it might find "salary" because their embeddings are close together.
• Cosine Similarity: A mathematical formula used to measure how similar two pieces of text are by looking at the angle between their vector embeddings.
• Large Language Model (LLM): We use TinyLlama-1.1B, a compact but powerful model that understands language and generates human-like responses.

---

✨ Features

• 📄 PDF Support: Upload any PDF document for instant analysis.
• 💬 Interactive Chat: Ask questions in natural language just like you're talking to a person.
• 🔍 Deep Retrieval: Performs semantic search to find the exact context needed.
• 🤖 Local AI: Designed to be lightweight, running efficiently on modern machines.
• 🖥 Sleek UI: A clean and minimal interface built with Streamlit for a premium user experience.
• 🔁 Contextual Memory: Remembers your previous questions for a smooth conversation flow.

---

🏗 System Architecture

The following diagram illustrates how your document flows through the system to become an answer:

graph TD
    A[User Uploads PDF] --> B[Text Extraction]
    B --> C[Text Chunking]
    C --> D[Embedding Generation]
    D --> E[In-Memory Vector Storage]
    
    F[User Asks Question] --> G[Query Embedding]
    G --> H[Semantic Search / Similarity Check]
    E -.-> H
    H --> I[Context Retrieval]
    I --> J[Prompt Augmentation]
    J --> K[LLM Generation]
    K --> L[Answer Displayed to User]

Loading

---

🔄 The Technical Workflow

1. Text Extraction: We use PyPDF to pull raw text from your PDF.
2. Chunking: To maintain accuracy, we split long text into smaller, overlapping chunks (approx. 500 characters each).
3. Vectorization: Each chunk is converted into a vector using the all-MiniLM-L6-v2 model.
4. Retrieval: When you ask a question, we find the top 5 most similar chunks using Cosine Similarity.
5. Generation: The top 5 chunks + your question are sent to TinyLlama to produce the final answer.

---

🧰 Technology Stack

Category	Technology Used	Description
Language	Python 3.x	The core programming language.
Frontend	Streamlit	Used for the interactive web interface.
Embeddings	SentenceTransformers	Specifically the all-MiniLM-L6-v2 model for vectorizing text.
LLM	TinyLlama-1.1B-Chat	The "brain" that generates human-like responses.
Math/Search	Scikit-learn / NumPy	For calculating cosine similarity and managing vector data.
PDF Handling	PyPDF	For extracting text from PDF files.
Chunking	LangChain	For intelligently splitting text into manageable pieces.

---

⚡ Installation & Setup

1. Clone the repository

git clone <repository-link>
cd <repository-name>

2. Create and Activate a Virtual Environment

Windows:

python -m venv venv
venv\Scripts\activate

Linux / Mac:

python -m venv venv
source venv/bin/activate

3. Install Required Dependencies

pip install -r requirements.txt

---

▶ Running the Application

Once the setup is complete, start the server using:

streamlit run app.py

The application will automatically open in your default web browser (usually at http://localhost:8501).

---

💡 Example Queries to Try

• "What is the main objective of this document?"
• "Can you list the key requirements mentioned?"
• "Summarize the conclusion in 3 bullet points."
• "What does the document say about the budget/stipend?"

---

🚀 Future Roadmap

• Support for Multiple Documents in a single session.
• Support for Docx and Text files.
• Integration with powerful external Vector Databases for large-scale storage.
• Streaming Responses for a faster "typing" effect.
• Support for higher-parameter models (like Llama-3 or Mistral).

---

📚 Conclusion

Queryo Docs demonstrates the power of combining traditional information retrieval with modern generative AI. It turns static documents into interactive knowledge bases, making information more accessible than ever before.

---

👨‍💻 Author

Mohd Saif Ansari
B.Tech Computer Science
Galgotias University