Adding Structured Document (Semantic Block Grouping) for Python Users

[jbdebard]

Python users currently receive a flat List[TextItem] per page with no semantic structure. The markdown_layout pipeline already classifies every projected line into semantic Block objects (Heading, Table, Figure, Paragraph, ListItem, CodeBlock, HorizontalRule, GridFallback) to produce the markdown output, but those Block objects are discarded after rendering. ProjectedLine.spans: Vec<TextItem> carries the source items for each line, but is also discarded.

Could you surface this classification per blocks to Python as ParsedPage.document_blocks: List[DocumentBlock], from which users can build a structured document with labeled, item-grouped blocks ?

[logan-markewich]

This has been added :) Exposed in the CLI + all bindings

https://developers.llamaindex.ai/liteparse/guides/extraction/#layout-blocks