Please sign up or log in to continue.

University of Zimbabwe Automated Course Portal (Warm-Up)

Automated UZ Course Portal built with PaddleOCR & ERNIE 4.0. Converts static PDFs into responsive, branded student websites with official eMhare styling.

Comment

Inspiration

As a Business Management & Computing student at the University of Zimbabwe, I noticed that many essential course materials (like the Programming II outline) are locked in static PDF formats. These are hard to read on mobile devices and difficult to search.

I wanted to create a solution that bridges the gap between Heritage (Old Formats) and Innovation (Digital Access). My goal was to build a system that automatically converts university documents into modern, responsive, and branded web portals.

What it does

This project is an Automated Course Portal Generator. It takes a raw PDF course outline and transforms it into a fully deployed website.

  • Extracts Content: It identifies course codes (BMI 201), instructor details (Mr. Atkins Mugota), and core values.
  • Generates Code: It writes clean HTML5 and CSS3 code.
  • Brands Automatically: It applies the official University of Zimbabwe "eMhare" color scheme (Navy Blue & Gold) to ensure the output looks official.

How we built it

The project uses a powerful AI pipeline:

  1. PaddleOCR-VL: I used this model to perform layout analysis on the source PDF, extracting text while preserving the hierarchical structure (Headers vs. Body text).
  2. ERNIE 4.0: The extracted data was fed into the ERNIE Large Language Model. ERNIE was tasked with generating semantic HTML structure and modern styling.
  3. Human-in-the-Loop Refinement: I applied "Frontend Engineering" principles to customize the CSS variables, ensuring the primary-dark and accent colors matched the University's official identity exactly.
  4. Deployment: The final solution is hosted on GitHub Pages for instant global access.

Challenges we ran into

The biggest challenge was Hallucination vs. Accuracy. Early attempts resulted in generic "template" styles. I learned that by refining the prompt and manually injecting specific CSS variables (like the specific #002855 eMhare Blue), I could force the AI to adhere to strict brand guidelines.

What we learned

I learned how to effectively chain Computer Vision (OCR) with Generative AI (LLM). This "Warm-Up" success has prepared me to build my next project: The Nhaka Intelligent Archivist, which will apply these same principles to restoring historical African documents.

Built With

Share this project:

Updates