PDF2Markdown.jl

Convert scanned and regular PDFs into structured Markdown using local LLMs (via Ollama) — 100% Julia, no cloud, no APIs, no nonsense.

OCR, layout parsing, and Markdown generation in one clean Julia pipeline.

---

Features

• Converts each page of a PDF to high-resolution PNG images using Poppler_jll
• Sends images to a local Ollama instance (e.g., gemma3:12b)
• 100% local — no internet required, your data stays on your machine
• Outputs structured, clean Markdown

---

Installation

using Pkg
Pkg.add(url="https://github.com/your-username/PDF2Markdown.jl")

Or clone the repo manually and Pkg.develop.

---

Usage

using PDF2Markdown

text = extract_text_from_pdf("path/to/your.pdf")
println(text)

You must have ollama installed and running locally, with a model like gemma3:12b pulled.

---

Requirements

• Julia ≥ 1.9
• Ollama running locally
• A pulled model like gemma3:12b or gemma3:4b
• Poppler must be functional via Poppler_jll

---

Example

text = extract_text_from_pdf("document.pdf")
write("output.md", text)

---

Notes

• Internally uses:
  • Poppler_jll to convert PDF pages to images
  • Base64 to encode images for Ollama
  • HTTP.jl and JSON3.jl to communicate with the LLM
• If you're converting large PDFs, consider batching pages.

---

Related Project

🔗 Python version: pdf2md-ollama

---

Credits

Inspired by this article on Medium

---

License

MIT