Extract PDF text and tables with AI
Recover reading order and tables from a mixed-layout report, with a readable document and a structured record of where each element appeared.
Extract PDF text and tables with AI when a report looks fine on screen but turns into a jumble when copied. Give pdf-parser a file that mixes columns, ruled tables and a few scanned pages. It first inspects the layout of each page so that the readable text can be extracted locally and the pages that need eyes can be set aside.
The Markdown output suits editing or quoting. The JSON output records the type, page and position of individual elements. A rendered image remains available for pages that need transcription. You can go back to the scan to check a questionable digit before importing the table into another document.
What it does
- Recover the column order in a research report.
- Extract a table and verify its row labels against the source page.
- Transcribe a scanned insert without applying OCR to the entire document.
- Identify the values visible in a chart and keep track of which figures were read visually.
- Check a footnote against the page image when extraction places it in the middle of a paragraph.
Install and use
- Use a file-capable agent with Python, PyMuPDF and pdfplumber. A vision-capable agent is needed for the visual fallback.
- Choose an input PDF and an output folder. Review the page triage and run the local extraction.
- Complete any flagged pages in both outputs. Check the important numbers against the original and look for unresolved placeholders.
FAQ
Source
Made by memtomem. Original skill: pdf-parser, published on GitHub under the MIT licence.
View on GitHub