Doctra is an open-source toolkit that converts PDFs into structured data by combining layout analysis, OCR, and Vision LMs (VLMs). It extracts text, tables, and charts/figures, then exports to Markdown, HTML, and Excel, with a CLI, Python API, and Gradio UI.
<ul>
<li>GitHub: <a href="https://github.com/AdemBoukhris457/Doctra" target="_blank" rel="noopener nofollow">https://github.com/AdemBoukhris457/Doctra</a></li>
<li>Docs: <a href="https://ademboukhris457.github.io/Doctra/index.html" target="_blank" rel="noopener nofollow">https://ademboukhris457.github.io/Doctra/index.html</a></li>
</ul>
<hr>
<h2>Why Doctra?</h2>
Most PDFs weren’t designed for data reuse. Doctra aims to make real-world documents (scans, multi-column layouts, noisy reports) immediately usable in data workflows, no manual copy/paste. For low-quality scans, page restoration can clean and enhance pages before OCR to boost accuracy.
<hr>
<h2>What it does</h2>
<ul>
<li>Layout detection → segment pages into blocks/regions for routing.</li>
<li>OCR → PaddleOCR and/or Tesseract for robust text on scans.</li>
<li>Vision LMs → interpret tables and charts into structured text/JSON.</li>
<li>Page restoration (optional) → denoise, deblur, and enhance scanned PDF pages to improve OCR/VLM results</li>
<li>Exports → Markdown, HTML, and Excel (with table sheets).</li>
<li>Interfaces → CLI for batch jobs, Python API for pipelines, Gradio UI for demos.</li>
<li>Providers → works with OpenAI, Gemini, Anthropic, OpenRouter (and more coming).</li>
</ul>
<hr>
<h2>Quick Start</h2>
<ol>
<li>Install with <code>pip install doctra</code>.</li>
<li>Run the CLI to parse a PDF and save outputs to a folder.</li>
<li>Or use the Python API to request Markdown/HTML/Excel outputs.
Full guides &amp; examples: <a href="https://ademboukhris457.github.io/Doctra/index.html" target="_blank" rel="noopener nofollow">https://ademboukhris457.github.io/Doctra/index.html</a></li>
</ol>
<hr>
<h2>Example Outputs</h2>
<ul>
<li>Tables → Markdown/HTML tables and an Excel file with one sheet per table.</li>
<li>Charts → human-readable descriptions + structured data when recoverable.</li>
<li>Text → clean Markdown/HTML with headings, paragraphs, and lists.</li>
</ul>
<hr>
<h2>Who is it for?</h2>
<ul>
<li>Data engineers turning reports into datasets</li>
<li>Researchers/analysts extracting tables/charts from papers</li>
<li>Back-office teams processing invoices, statements, contracts</li>
<li>Builders who need a simple CLI/UI to test and a Python API to integrate</li>
</ul>
<hr>
<h2>Contribute / Feedback</h2>
Issues and PRs are welcome. Share tricky PDFs (scans, multi-column, complex tables) to help improve robustness.
<ul>
<li>Repo: <a href="https://github.com/AdemBoukhris457/Doctra" target="_blank" rel="noopener nofollow">https://github.com/AdemBoukhris457/Doctra</a></li>
<li>Docs: <a href="https://ademboukhris457.github.io/Doctra/index.html" target="_blank" rel="noopener nofollow">https://ademboukhris457.github.io/Doctra/index.html</a></li>
<li>⭐ If Doctra helps you, consider starring the repo!</li>
</ul>

Squad for devs building apps with GenAI and LLMs! Connect, share, exchange ideas, and get feedback. 

Shipping Doctra: an open-source document parsing toolkit