Docling — Document Parsing for AI Pipelines

Supported Formats

Table Extraction

Docling accurately extracts tables preserving structure:

result = converter.convert("financial_report.pdf")
for table in result.document.tables:
    print(table.export_to_dataframe())  # pandas DataFrame

Figure Handling

Figures are extracted with captions and alt-text:

for figure in result.document.pictures:
    print(f"Caption: {figure.caption}")
    figure.image.save(f"figure_{figure.id}.png")

OCR for Scanned Documents

Built-in OCR for scanned PDFs and images:

converter = DocumentConverter(ocr=True)
result = converter.convert("scanned_document.pdf")

Batch Processing

from docling.document_converter import DocumentConverter
from pathlib import Path

converter = DocumentConverter()
docs = Path("./documents").glob("*.pdf")
for doc in docs:
    result = converter.convert(str(doc))
    markdown = result.document.export_to_markdown()
    Path(f"./output/{doc.stem}.md").write_text(markdown)

Integration with RAG

# With Haystack
from docling.integrations.haystack import DoclingConverter
converter = DoclingConverter()
documents = converter.run(sources=["report.pdf"])

# With LangChain
from docling.integrations.langchain import DoclingLoader
loader = DoclingLoader(file_path="report.pdf")
docs = loader.load()

Key Stats

• 18,000+ GitHub stars
• By IBM Research
• 6 input formats supported
• Table, figure, equation extraction
• OCR for scanned documents
• Haystack, LangChain integrations

FAQ

Q: What is Docling? A: Docling is a document parsing library by IBM Research that converts PDF, DOCX, PPTX, and other formats into clean Markdown or JSON for LLM consumption, handling tables, figures, and complex layouts.

Q: Is Docling free? A: Yes, open-source under MIT license by IBM Research.

Q: How accurate is Docling compared to PyPDF? A: Significantly more accurate for complex documents. Docling uses deep learning models to understand document layout, while PyPDF does simple text extraction.