PaperFlow

PaperFlow is an open-source post-processing layer for PDF-to-Markdown workflows.

It does not try to be the parser. It takes raw markdown from a parser you choose and upgrades it into structured, knowledge-ready markdown with:

• normalized LaTeX delimiters
• linked citations as standard footnotes
• figure and table jump links
• YAML frontmatter
• cleaned repeated headers and footers

PyPI package: https://pypi.org/project/paperflow-postprocess/
GitHub: https://github.com/TylerMorrison21/paperflow
Project page: https://www.paperflowing.com

Install

pip install paperflow-postprocess

Quick Start

from paperflow_postprocess import enhance

raw_markdown = """
Text with citation [1].

## References

[1] Example Author. Example Paper.
"""

markdown = enhance(
    raw_markdown=raw_markdown,
    images={},
    metadata={
        "title": "Example Paper",
        "authors": ["Example Author"],
        "source": "https://example.com/paper",
        "date": "2026-03-11",
    },
)

print(markdown)

PDF Parsers - bring your own

PaperFlow is a post-processing layer. It enhances raw Markdown from any upstream PDF parser. You need to choose a parser:

Option 1: Datalab Marker API (recommended, easiest)

• Sign up at datalab.to - $25/month free credits
• Cloud API, no GPU needed
• Set DATALAB_API_KEY in your .env
• PaperFlow's built-in api/services/marker.py calls this automatically

Option 2: Marker (self-hosted, free)

• github.com/datalab-to/marker
• Run locally with GPU (CUDA) or CPU
• Free for orgs under $5M revenue
• You'll need to modify api/services/marker.py to call your local endpoint

Option 3: MinerU (self-hosted, free)

• github.com/opendatalab/MinerU
• Strong on Chinese docs, scientific papers, complex tables
• Outputs Markdown + LaTeX - compatible with PaperFlow's postprocess
• Needs GPU (~6GB VRAM minimum)
• Replace marker.py with a MinerU client

Option 4: Docling, PyMuPDF4LLM, or any other parser

• Any tool that outputs Markdown will work
• Feed the raw Markdown into paperflow_postprocess.enhance() or save it and run it through the API pipeline

Using the pip package with any parser

from paperflow_postprocess import enhance

# Get raw markdown from ANY parser
raw_md = my_parser.convert("paper.pdf")

# Enhance with PaperFlow
result = enhance(raw_md, images={}, metadata={"title": "My Paper"})

# result has footnotes, fixed LaTeX, figure links, YAML frontmatter

The whole point of PaperFlow is that parsing is commoditized. Marker, MinerU, LlamaParse, Docling, and PyMuPDF4LLM can all produce decent raw output. The post-processing layer is where the value is, and that is what PaperFlow does.

Visual Comparison

Source PDF	Generic converter output	PaperFlow output

Additional example:

What enhance() does

enhance() upgrades parser output into structured markdown with:

• standard footnotes [^N] from inline citations like [1], [1, 2], or [1-3]
• normalized LaTeX delimiters using $...$ and $$...$$
• figure links like [[#^fig-3|Fig. 3]]
• table links like [[#^tab-2|Table 2]]
• YAML frontmatter with title, authors, source, date, and extraction hash
• cleaned repeated headers, footers, and page number lines

API

• enhance(raw_markdown, images=None, metadata=None)
• postprocess(raw_markdown, images=None, metadata=None) for backward compatibility
• fix_latex_delimiters(md)
• clean_headers_footers(md)
• convert_to_footnotes(md)
• linkify_figures(md)
• linkify_tables(md)
• inject_frontmatter(md, metadata)