Transforming scanned documents into structured, searchable knowledge using Optical Character Recognition, intelligent table extraction, and MongoDB.
Developed by Ankit Kumar Thakur
B.Tech Information Technology Engineering
Maharaja Agrasen Institute of Technology, Delhi
DocuQuery-AI is an intelligent document processing system designed to automate the extraction, organization, storage, and retrieval of information from scanned documents.
Traditional document management systems often require manual data entry, making them time-consuming, error-prone, and difficult to scale. DocuQuery-AI addresses this challenge by integrating Optical Character Recognition (OCR), document preprocessing, intelligent table extraction, structured database storage, and keyword-based information retrieval into a unified web application.
The project demonstrates an end-to-end document intelligence pipeline capable of processing printed and handwritten documents while significantly reducing manual effort involved in digitization.
Every organization generates thousands of documents ranging from reports and invoices to handwritten forms and tabular records. Although these documents contain valuable information, much of it remains inaccessible because it exists only as scanned images or PDFs.
The motivation behind DocuQuery-AI is to bridge the gap between physical documents and searchable digital knowledge by developing a system capable of:
- Extracting textual information from scanned documents.
- Preserving tabular structures during extraction.
- Organizing extracted information into structured formats.
- Storing the extracted data efficiently.
- Allowing users to retrieve information instantly through keyword-based search.
- Intelligent OCR using PaddleOCR
- Printed and handwritten document processing
- Automatic table extraction
- Image preprocessing for improved recognition
- MongoDB integration for structured document storage
- Keyword-based document retrieval
- Flask-based interactive web interface
- Modular and extensible architecture
- Support for image and scanned document processing
- Structured JSON generation from extracted data
The core OCR pipeline of DocuQuery-AI is built around PaddleOCR, an industrial-grade Optical Character Recognition framework developed by Baidu.
PaddleOCR was selected because it provides:
- High text recognition accuracy
- Fast inference speed
- Robust multilingual support
- Excellent document layout understanding
- Reliable table recognition capabilities
- Lightweight deployment
Within this project, PaddleOCR performs:
- Text region detection
- Character recognition
- Text orientation correction
- Table content extraction
- Structured document understanding
Image preprocessing is carried out using OpenCV, improving document readability before OCR inference.
The extracted information is then cleaned, structured, and converted into JSON before being stored inside MongoDB.
During the development and research phase, multiple OCR methodologies were explored to understand their strengths, limitations, and suitability for intelligent document processing.
These include:
- PaddleOCR
- Microsoft TrOCR
- DocTR (Document Text Recognition)
- Traditional OCR pipelines
The current implementation uses PaddleOCR as the primary OCR engine due to its balance of accuracy, inference speed, and structured document support. However, the modular design allows future benchmarking and integration of transformer-based OCR models such as TrOCR and DocTR.
| Technology | Purpose |
|---|---|
| Python | Backend Development |
| Flask | Web Framework |
| PaddleOCR | Optical Character Recognition |
| OpenCV | Image Processing |
| MongoDB | NoSQL Database |
| PyMongo | MongoDB Integration |
| NumPy | Numerical Computing |
| Pandas | Data Processing |
| HTML | User Interface |
| CSS | Styling |
| JavaScript | Client-side Interaction |
Document Upload
│
▼
Image Preprocessing
│
▼
PaddleOCR
(Text Detection + Recognition)
│
▼
Table Extraction
│
▼
Data Cleaning
│
▼
JSON Structuring
│
▼
MongoDB Storage
│
▼
Keyword-Based Retrieval
│
▼
Relevant Search Results
User Uploads Document
│
▼
Flask Application
│
▼
Image Processing
(OpenCV)
│
▼
PaddleOCR
│
▼
Text & Table Extraction
│
▼
Structured JSON Output
│
▼
MongoDB Database
│
▼
Keyword Query Engine
│
▼
Search Results Display
The repository includes several representative documents used during development and testing.
These include:
- Printed tabular documents
- Handwritten tables
- Online handwritten samples
- Relation diagrams
- Sample records
- OCR benchmark images
These files are available in the sample_inputs directory.
Clone the repository
git clone https://github.com/NeuralImprint/DocuQuery-AI.gitMove into the project directory
cd DocuQuery-AICreate a virtual environment
conda create -n docuquery python=3.10Activate the environment
conda activate docuqueryInstall dependencies
pip install -r requirements.txtRun the application
python app.pyDocuQuery-AI
│
├── modules/
│ ├── hand_only_query.py
│ ├── knowledgegraph.py
│ ├── mongo_knowledgegraph.py
│ ├── mongosetup.py
│ ├── paddlemongo.py
│ └── queryextractor.py
│
├── sample_inputs/
│
├── static/
│
├── templates/
│
├── uploads/
│
├── app.py
├── requirements.txt
├── Procfile
└── README.md
DocuQuery-AI can be applied across multiple domains, including:
- Government record digitization
- Defence documentation
- Healthcare records
- Financial document management
- Educational archives
- Enterprise document management
- Legal documentation
- Research data organization
The modular architecture of DocuQuery-AI allows future expansion in several directions.
Potential enhancements include:
- Retrieval-Augmented Generation (RAG)
- Semantic document search
- Knowledge Graph construction
- Named Entity Recognition
- Vector database integration
- Transformer-based OCR benchmarking
- Multilingual document understanding
- Large Language Model integration
- Cloud-native deployment
- Automatic document classification
Ankit Kumar Thakur
B.Tech Information Technology Engineering
Maharaja Agrasen Institute of Technology, Delhi
Areas of Interest
- Artificial Intelligence
- Machine Learning
- Computer Vision
- Document Intelligence
- OCR Systems
- Deep Learning
- Generative AI
- Research & Development
GitHub: https://github.com/NeuralImprint
LinkedIn: https://www.linkedin.com/in/ankit-thakur-179a23237/
Copyright © 2026 Ankit Kumar Thakur.
This project is protected under a custom All Rights Reserved license.
The source code is provided for portfolio demonstration, academic reference, and research purposes only.
No part of this repository may be copied, modified, redistributed, or used in any form without prior written permission from the author.