One call for every format

Inside a DoclingDocument

Getting the text back out

PDFs

More than one file

Chunking for retrieval

Learn Docling from scratch: one call that converts Markdown, HTML, CSV and PDF into the same document object, the exports that get text back out, the PDF pipeline, and chunking for retrieval. No API key, and a searchable handbook at the end.

Docling

Where the course goes

Docling tutorial: turn any document into text you can search

What you are going to build

Reading Markdown, HTML and CSV without Docling

Three files, three parsers

Docling convert(): one call for Markdown, HTML and CSV

One call for all three

Docling ConversionResult: status, input and errors

What comes back from convert

iterate_items(): inspect a Docling document piece by piece

Seeing what Docling saw

DoclingDocument: texts, tables, pictures and groups

The lists a document is made of

Docling labels: title, section_header, text and list_item

What Docling decided each piece is

Docling reading order: body, children and parents

The order the document is meant to be read in

Docling tables: export_to_dataframe and the cell grid

A table you can calculate with

Docling save_as_json: store a converted document losslessly

Putting a document away and getting it back

Docling export_to_markdown: labels, ranges and compact tables

Markdown, and the arguments that change it

export_to_text and export_to_html in Docling

Plain text, and HTML that keeps the spans

Docling DocTags: the model-facing document format

DocTags, the format the models speak

Convert a PDF with Docling: the models and the first run

Converting a PDF

Docling provenance: page numbers and bounding boxes

Which page, and where on it

PdfPipelineOptions in Docling: OCR, tables and page images

The switches on the PDF pipeline

Docling heading levels: recovering PDF outline depth

The heading levels a PDF does not have

Docling confidence grades: spotting a bad conversion

How sure the conversion is

Docling convert_all: batch conversion and its failures

Four files in one call

Docling allowed_formats and unsupported file errors

The files Docling will not read

Docling DocumentStream: convert an upload without saving it

Converting bytes that were never a file

Docling CLI: convert files from the command line

Docling from the command line

HierarchicalChunker in Docling: one chunk per element

Cutting a document into pieces

Docling contextualize(): headings attached to every chunk

The headings a chunk carries with it

Docling HybridChunker: token limits and merged chunks

Chunks that fit

Docling serializers: control the text a chunk contains

Changing how a table reads

Chunking for retrieval quiz

A chunk of two bullet points about tracking codes never matches a search for <em>shipping</em>. What do you index instead of <code>chunk.text</code>?