Workflow leaderboard / Research / PDF & document processing

Convert scholarly PDFs to auditable Markdown

将学术 PDF 转为可审计 Markdown

Research / 科研 PDF & document processing / 文献处理 curated verified 2026-08-02

Extract born-digital or scanned papers into structured Markdown while retaining page-level artifacts for verification.

Variants

Free & local / 免费本地 ~20 min setup

16 GB RAM is recommended. OCR and table extraction are faster with a supported GPU but do not require one.

ComponentRolePriceLink
Docling 2.117.0 PDF layout, table and OCR conversion Free / 免费MIT-licensed local software; model downloads and compute are separatechecked 2026-08-02 https://github.com/docling-project/docling/releases/tag/v2.117.0

Convert and preserve an audit bundle

uv venv --python 3.12 .venv
. .venv/bin/activate
uv pip install 'docling==2.117.0'
mkdir -p converted/source converted/output
cp paper.pdf converted/source/
docling converted/source/paper.pdf --to md --to json --output converted/output
sha256sum converted/source/paper.pdf converted/output/* > converted/SHA256SUMS

Page-grounded extraction review

Compare the converted Markdown and JSON with the source PDF page by page. Report missing headings, merged columns, broken equations, lost footnotes, incorrect reading order and OCR uncertainty. Quote only the minimum text needed to locate each defect and give the PDF page number. Do not repair content from general knowledge.

Known pitfalls

  • Reading order can be wrong in multi-column layouts.
  • Equations, footnotes and spanning tables require visual comparison with the PDF.
  • OCR may produce plausible but incorrect numbers.
  • Do not discard the original PDF or page references after conversion.

Evidence