📄 Query Retrieval System

A powerful multi-format document question-answering system that extracts, processes, and searches through various document types to provide contextually accurate answers using semantic search and LLMs.

🚀 Overview

This system takes documents (via URL or file), processes them through a smart extraction pipeline (with OCR for scanned content), splits the content into semantic chunks, embeds them into vectors, and retrieves the most relevant chunks for a given question.
It then sends the context to an LLM (GPT-4.1/GPT-5.1) to produce accurate, document-based answers.

✨ Key Features

• Multi-format Support: PDF, DOCX, PPTX, XLSX, ZIP archives, and images.
• Smart Extraction Pipeline:
  • MIME type detection
  • Format-specific extractors
  • OCR for scanned documents & images
  • Fallback conversion for unknown formats
• Parallel Processing:
  • Concurrent document extraction
  • Parallel question processing
• Semantic Search:
  • Embedding-based similarity search
  • Retrieves relevant chunks even without exact keyword matches
• Scalable LLM Integration:
  • Key rotation for LLM API calls
  • JSON-formatted output for easy parsing

🛠️ Tech Stack

• Backend: Node.js + Express
• AI / ML:
  • Xenova Transformers – Embedding generation
  • GPT-4.1 via GitHub Models API – Answer generation
• OCR: Tesseract.js
• Document Extraction:
  • pdf-parse – PDF text extraction
  • mammoth – DOCX text extraction
  • unzipit – PPTX/DOCX XML parsing
  • xlsx – Excel parsing
  • libreoffice-convert – Format conversion
• Utilities:
  • axios, fs-extra, uuid, file-type, p-limit, sharp, canvas

📦 Installation

1️⃣ Clone the repository

git clone https://github.com/DeepakSingh0027/Query-Retrieval-System.git

2️⃣ Install dependencies

npm install

3️⃣ Post-install rebuild (required for sharp & canvas)

npm rebuild sharp && npm rebuild canvas

4️⃣ Install LibreOffice (Required for libreoffice-convert)

Ubuntu/Debian

sudo apt update
sudo apt install libreoffice

MacOS (Homebrew)

brew install --cask libreoffice

Windows

-Download from LibreOffice Official Website.:https://www.libreoffice.org/download/download/
-Add LibreOffice's program folder path (e.g., C:\Program Files\LibreOffice\program) to your system PATH.

5️⃣ Environment Variables

Create a .env file in the server directory: (Note: Already pushed.)

GITHUB_TOKEN=<your_main_github_api_token>
GITHUB_TOKEN2=<token_2>
GITHUB_TOKEN3=<token_3>
.. up to GITHUB_TOKEN12 if needed

▶️ Running the Server

cd server
node server.js

📂 File Structure

server/
│── server.js # Main entry point
│── controllers/
│ └── hackrxController.js # Main processing workflow
│── utils/
│ ├── chunkText.js # Chunking logic
│ ├── semanticSearch.js # Embedding & similarity search
│ ├── queryModel.js # LLM integration
│ └── functions/
│ ├── extractor.js # File handling & extraction
│ └── extractor2.js # PDF conversion fallback
│── extractorsUtils/
│ ├── pptExtract.js
│ ├── docxExtract.js
│ └── xlExtract.js
│── package.json
│── .env
(Note: There are more files that supports/helps the main workflow that is not mentioned here.)

📜 Example Workflow

1 - User sends API request with:
documents: array of URLs or file paths
questions: array of strings
2 - System:
Detects file type → extracts text (with OCR if needed)
Splits text into semantic chunks
Embeds chunks into vectors
Finds top relevant chunks per question
Sends context to LLM
LLM returns JSON-formatted answers
All answers were merged and send back with JSON response.