DocWeb

• 

  Homepage

• 

  Preview page

• 

  Generated HTML page

Inspiration

Many organizations struggle with converting static PDF documents into accessible, interactive web content. The rise of AI models like Baidu's ERNIE and PaddleOCR-VL provided the perfect opportunity to build an intelligent, automated solution.

What it does

DocWeb is a web application that transforms PDF documents into responsive webpages using artificial intelligence. The platform:

• Extracts text from multi-page PDFs using PaddleOCR-VL for accurate optical character recognition
• Converts content into structured Markdown format
• Generates HTML webpages with AI-powered styling via ERNIE 4.5
• Exports multiple formats (HTML, Markdown, JSON) for maximum flexibility
• Provides real-time previews of generated content

How I built it

DocWeb was built using a modern, modular architecture:

• Frontend: Streamlit for an intuitive, responsive user interface
• OCR Engine: PaddleOCR-VL for document text extraction
• AI Models: Baidu's ERNIE 4.5 for intelligent HTML generation and styling
• Processing Pipeline: Custom Python modules for PDF extraction, Markdown conversion, and HTML generation
• Styling: CSS-based theming for a clean, professional interface with full customization

The workflow follows a logical five-step process: Upload → Extract → Convert → Generate → Download

Challenges we ran into

Several technical challenges shaped our development:

• OCR Accuracy: Ensuring reliable text extraction from PDFs with varying quality, layouts, and fonts
• Markdown Conversion: Preserving document structure and formatting during conversion
• UI/UX Design: Creating an intuitive interface while maintaining performance with large files

Accomplishments that we're proud of

• Multiple Export Formats: Users can download HTML, Markdown, and JSON—maximizing flexibility
• Real-time Preview: Implemented live preview functionality so users see results instantly
• Error Handling: Robust error management with user-friendly feedback messages

What I learned

This project taught me valuable lessons:

• How to integrate multiple AI services (OCR + LLM) into a cohesive workflow
• The balance between automation and user control in AI applications
• Document processing challenges: PDFs are not uniformly structured, requiring flexible approaches
• The power of combining multiple specialized AI models for superior results

What's next for DocWeb

Future enhancements we're planning:

• Batch Processing: Enable users to convert multiple PDFs simultaneously
• Advanced Styling: Give users templates and customization options for generated HTML
• API Endpoint: Expose DocWeb as an API for enterprise integration

Built With

• ernie
• paddleocr-vl
• streamlit