A slide is a page
The slide number is the citation. A sentence without a slide number cannot be checked against the deck.
Read the title first, then the body in a stable order. A pile of text-box contents is not a slide.
Notes and pictures
Speaker notes are not the slide body. The current PPTX path does not read them, so a missing notes block means the notes were not extracted.
A picture on a slide is stored as an image asset without recognized text. Image recognition is a paid-workspace capability, and the job route says whether that path ran.
Chunk by slide
One slide, or a title plus its body, is usually the right chunk. Merging ten slides into one window mixes unrelated claims.
Store the slide index on the chunk. The DocIR page list is that index.
What the PPTX path writes
A PPTX job records route office-ooxml and is parsed in the container. Each slide is one page. The first block is a level-1 heading whose text is Slide N, with N starting at 1. Shapes are then read in top-to-left order. A text frame becomes a paragraph. A shape table becomes a table block. A picture becomes an image asset and does not add recognized words.
Speaker notes are not read. Do not invent a notes block when the field is absent. Legacy .ppt files need LibreOffice conversion before this path can run.
One text frame becomes one paragraph block, including when the frame holds several lines. An empty text frame is skipped. A shape that is not a table or a text frame is stored as an image asset and adds no words. Shapes are ordered by top, then by left, and a kept shape can include a box in points. The first block is the Slide N heading. Cite that heading, which starts at 1, and keep pageIndex 0 for the first slide. A legacy .ppt file fails closed when LibreOffice is unavailable, with no partial slide text.
- Cite Slide N from the heading, which is already 1-based.
- pageIndex 0 is the first slide.
- Chunk one slide, using the blocks on that page.
- Treat a picture as an asset until an image route is allowed.
# Slide 1
Revenue by segment
The file to send
Send one PPTX. result.manifest.decision.route is office-ooxml. Each slide is one DocIR page. The first block is a level-1 heading, Slide N, with N starting at 1. pageIndex 0 is the first slide.
Speaker notes are not read. A picture is an image asset and adds no recognized words. One text frame becomes one paragraph. A legacy .ppt fails closed when LibreOffice is unavailable.
- Cite Slide N from the heading.
- Read shapes in the top-then-left order the page lists.
- Chunk one slide, using the blocks on that page.
export DOCPARSE_URL="https://docparse.genedai.me"
export DOCPARSE_API_KEY="your_key"
curl -fsS -X POST \
"$DOCPARSE_URL/v1/parse?wait=true" \
-H "Authorization: Bearer $DOCPARSE_API_KEY" \
-F "file=@deck.pptx" \
-F "mode=both" \
-o result.json
jq -r '.result.markdown' result.json
jq '.result.document.schemaVersion, .result.document.pages[0].blockIds' result.json
jq '.result.manifest.decision.route, .result.manifest.parserVersion' result.json
Frequently asked questions
Questions this guide answers
Does DocParse parse PPTX?
Yes. PPTX is routed through POST /v1/parse. The job route is office-ooxml. The result is Markdown and docir.v1, with one DocIR page per slide.
Should speaker notes be mixed into the slide text?
Notes are a different source from the slide. The current PPTX path does not extract speaker notes, so the Markdown contains only the slide shapes that were read.
How should a deck be chunked?
Use one slide as the chunk. The heading Slide N is 1-based, and the DocIR page index for that slide is N minus 1. One text frame is one paragraph block. Empty text frames are skipped. Speaker notes are not in the Markdown.