knowledge-extractor-ai 0.1.0

Read any PDF and preserve its document structure (headings, tables, lists, key-value pairs, checkboxes, reading order, bounding boxes) as JSON. Powered by Docling.

knowledge-extractor-ai

Read any PDF and preserve its document structure — headings, paragraphs, tables, lists, key-value pairs, checkboxes, reading order and bounding boxes — as clean JSON, instead of flattening it to plain text.

Powered by Docling for layout analysis.

Install

pip install knowledge-extractor-ai

Usage

Python

from knowledgeextract.pdf import extract_pdf

doc = extract_pdf("myfile.pdf")
doc.to_json("result.json")        # write structured JSON
print(doc.markdown)               # Markdown view

Command line

pdf-extract extract myfile.pdf -o result.json -m result.md

The first run downloads/loads Docling's layout models and may take a few minutes; subsequent runs are fast.

What gets extracted

Field	Notes
Headings	with level
Paragraphs	body text blocks
Tables	rows + dimensions
Lists	items grouped, ordered/unordered
Checkboxes	checked state + label
Key-value pairs	label/value regions
Images	reference + bounding box (no bytes)
Bounding boxes	on every element (PDF points, top-left origin)
Reading order	document-wide ordering
Markdown	rendered view of the whole document

Output shape

{
  "metadata": { "filename": "myfile.pdf", "page_count": 1, "engine": "docling" },
  "pages": [
    {
      "page_number": 1,
      "width": 612.0,
      "height": 792.0,
      "elements": [
        { "type": "heading", "order": 0, "page_number": 1, "level": 1,
          "bbox": { "x0": 78.0, "y0": 111.0, "x1": 178.0, "y1": 127.7 },
          "text": "1. Overview" }
      ]
    }
  ],
  "reading_order": [0, 1, 2],
  "markdown": "## 1. Overview\n..."
}

Project layout

Each file format is a self-contained subpackage. PDF ships today; Word, Excel and others plug in alongside it.

knowledgeextract/
└── pdf/
    ├── converter.py   # Docling wrapper
    ├── normalize.py   # Docling -> schema
    ├── models.py      # JSON schema (Pydantic)
    └── cli.py         # pdf-extract command

License

See LICENSE.