From folder to ConversionResult

Lesson 1 built a DocumentConverter but never fed it anything. This lesson does that, on sample_data/quarterly_report.pdf, a two-page PDF with headings, a table, and a chart image, generated for this course and used again in several later lessons.

converter = DocumentConverter()
result = converter.convert(SAMPLE_DATA / "quarterly_report.pdf")

convert() accepts a path, a URL string, or a DocumentStream for in-memory bytes (that last one shows up in Lesson 9's OCR work). It returns a ConversionResult, not a document directly, that result carries a .status (did the conversion succeed), .input (what format was detected), .pages (per-page metadata), and .document, the actual DoclingDocument.

export_to_markdown()

result.document is a DoclingDocument, docling's unified in-memory representation, texts, tables, and pictures as structured objects, not a string. export_to_markdown() walks that structure and serializes it: headings become #/##, a detected table becomes a Markdown pipe table, a detected picture becomes an ![]() placeholder. This is the method you'll reach for most often in this course, it's the fastest path from "a document" to "text an LLM can read".

The first run of this lesson on a fresh machine takes noticeably longer, docling downloads its layout and table-structure model weights the first time they're used, then caches them locally for every run after that.

Checkpoint

  • convert(): takes a path, URL, or DocumentStream, returns a ConversionResult.
  • ConversionResult: .status, .input, .pages, .document.
  • .document: a DoclingDocument, docling's structured representation, not a plain string.
  • export_to_markdown(): serializes that structure to Markdown, headings, tables, and picture placeholders included.

If anything here still feels unclear, ask before moving to Lesson 3.