Newspaper-OCR-LLM

Uses LLM to iterate through a collection of scanned pages of historical multi-column newspapers and OCR the output to extract: articles and headline, along with some additional metadata. It returns a collection of sticles identifies on each page in JSON format.

---

What it does

• Reads scanned page images (JPG) from an input directory
• Sends them to an OpenAI model (e.g. o4-mini-2025-04-16) with a structured prompt optimised externally on OpenAI Playground for layout‑aware OCR
• Returns per‑page JSON with:
  • articles[] (title/headline, content/article body)
  • page/issue metadata
• Writes outputs to an output directory (one JSON file per scanned page image)

---

How to run

1. Regiter and get an OpenAI API Key;
2. Configure input and output folders;
3. Add scanned page images as JPG;
4. Run the Cell on Jupyter notebook.

---