From PDF to Live Web Page

Upload a PDF (resume, article, or guide), extract structured content with PaddleOCR-VL, convert it to Markdown, then use ERNIE to generate a responsive web page automatically.

Comment

Inspiration

Many valuable documents still live as static PDFs, which limits how easily their content can be reused on the web. I wanted to explore how modern multimodal AI can bridge this gap by automatically transforming PDFs into clean, readable web pages with minimal manual effort.

This project was inspired by the challenge of preserving document structure and meaning while making content more accessible and web-friendly.

What it does

This project converts a PDF into a live web page through an automated pipeline:

  1. PaddleOCR-VL extracts the text and layout information from the PDF.
  2. The extracted content is converted into structured Markdown.
  3. ERNIE generates a responsive HTML web page from the Markdown.
  4. The final page is deployed publicly using GitHub Pages.

The result is a clean, readable web page generated directly from a PDF source.

How I built it

  • Used PaddleOCR-VL to perform OCR and layout extraction from the PDF.
  • Transformed the OCR output into structured Markdown.
  • Used ERNIE to generate semantic HTML and page structure.
  • Deployed the generated page using GitHub Pages for public access.

This approach demonstrates how OCR and large language models can work together to automate document-to-web workflows.

Challenges I ran into

  • Preserving layout hierarchy when converting OCR output into Markdown.
  • Ensuring the generated HTML remained readable and well-structured.
  • Balancing automation with simplicity to keep the pipeline reliable.

Each challenge was addressed by keeping the workflow modular and validating each step before moving to the next.

What I learned

  • How PaddleOCR-VL captures both text and layout context from PDFs.
  • How ERNIE can transform structured content into meaningful web layouts.
  • Best practices for deploying lightweight AI-generated content using GitHub Pages.

This project highlights the potential of combining OCR and language models to make static documents more accessible and reusable.

Built With

Share this project:

Updates