Web Builder – PDF to Website Generator

Automatically convert a PDF document into a live website using OCR, Markdown transformation, AI-powered HTML generation, and GitHub Pages auto-deployment.

Comment

🧠 Web Builder – PDF to Website Generator

Automatically convert a PDF document into a live website using OCR, Markdown transformation, AI-powered HTML generation, and GitHub Pages auto-deployment.

This project demonstrates end-to-end backend architecture, AI integration, rate-limit resilience, and production-grade system design.

🚀 What This Project Does

  1. Upload a PDF file
  2. Extract text using OCR (Python + PaddleOCR)
  3. Convert OCR output to Markdown
  4. Convert Markdown to HTML (AI with fallback)
  5. Save index.html
  6. Auto-deploy to GitHub Pages

👉 Result: A publicly accessible website generated from a PDF.

🏗 High-Level Architecture

Client (Browser) | Spring Boot Backend (Java) | +--> Python OCR Service (FastAPI + PaddleOCR) | +--> Markdown Generator | +--> AI HTML Generator (OpenAI) | | | +--> Local HTML Fallback (if AI fails) | +--> Save index.html | +--> GitHub Pages Deployment

📁 Project Structure

ocr-project/ ├── java-backend/ │ ├── controller/ │ │ └── PdfController.java │ ├── service/ │ │ ├── OcrClient.java │ │ ├── MarkdownService.java │ │ ├── AiHtmlService.java │ │ ├── HtmlSaveService.java │ │ └── GithubDeployService.java │ └── application.yml │ ├── python-ocr/ │ └── ocr_server.py │ └── site/ └── index.html # Generated output

🔄 End-to-End Flow

1️⃣ Upload PDF

POST /upload multipart/form-data

Handled by PdfController.

2️⃣ OCR (Python Service)

  • Built using FastAPI
  • Uses PaddleOCR
  • Java backend sends PDF bytes
  • Python service returns structured OCR JSON

3️⃣ OCR → Markdown

Text blocks are converted into readable Markdown:

  • Headings
  • Paragraphs
  • Lists

4️⃣ Markdown → HTML (AI + Fallback)

Handled by AiHtmlService.

Primary (AI)

  • OpenAI Chat Completions API
  • Strict rate-limit handling:
    • Semaphore (1 request at a time)
    • Hard throttling
    • Exponential backoff retries

Fallback (Guaranteed)

If OpenAI is unavailable (429 / quota / network):

  • Local HTML renderer is used
  • Pipeline never fails

This makes the system production-safe.

5️⃣ Save HTML

site/index.html

6️⃣ GitHub Pages Deployment

  • Uses JGit
  • Auto-commit & push to GitHub repository
  • GitHub Pages serves the site

Live URL: https://.github.io//

⚠️ Failure & Rate-Limit Handling

This system is designed to never break.

Scenario Behavior
OpenAI rate-limited (429) Fallback HTML
OpenAI down Fallback HTML
Invalid OCR output Fallback HTML
No API key Fallback HTML

✔ Website is always generated.

🛠 Setup Instructions

Prerequisites

  • Java 17+
  • Python 3.10+
  • Git
  • GitHub account
  • (Optional) OpenAI API key

Environment Variables

Java Backend


OPENAI_API_KEY=sk-xxxxx        # optional
GITHUB_TOKEN=ghp_xxxxxx

Start Python OCR Service


cd python-ocr
python ocr_server.py
Runs on: http://localhost:8000

Start Java Backend


cd java-backend
mvn spring-boot:run
Runs on: http://localhost:8080

Test the System


Upload a PDF via:
POST http://localhost:8080/upload

Built With

Share this project:

Updates

Submission history