PPTX Parsing API

Presentations Are Visual Containers That Defeat Text Extraction

A .pptx file is a zip archive containing PresentationML — XML that encodes slides, shapes, text frames, tables, and speaker notes. Most parsers render slides to images or flatten them to plain text, losing the distinction between title, body, notes, and table content.

AILANG Parse reads PresentationML directly. Each slide becomes a section block with typed content: titles, body text, tables, and speaker notes are all separate blocks. The parser extracts from shape placeholders in layout order, so content appears in reading sequence.

Raw PresentationML vs Structured Output

<p:sld>
  <p:cSld>
    <p:spTree>
      <p:sp>
        <p:nvSpPr><p:nvPr>
          <p:ph type="title"/>
        </p:nvPr></p:nvSpPr>
        <p:txBody>
          <a:p><a:r>
            <a:t>Q1 Results</a:t>
          </a:r></a:p>
        </p:txBody>
      </p:sp>
      <p:sp>
        <p:nvSpPr><p:nvPr>
          <p:ph idx="1"/>
        </p:nvPr></p:nvSpPr>
        <p:txBody>
          <a:p><a:r>
            <a:t>Revenue up 15% YoY</a:t>
          </a:r></a:p>
          <a:p><a:r>
            <a:t>New markets: APAC, LATAM</a:t>
          </a:r></a:p>
        </p:txBody>
      </p:sp>
    </p:spTree>
  </p:cSld>
</p:sld>

<!-- notesSlide1.xml -->
<p:notes>
  <p:cSld>
    <p:spTree>
      <p:sp><p:txBody>
        <a:p><a:r>
          <a:t>Mention APAC expansion
timeline: Q3 launch</a:t>
        </a:r></a:p>
      </p:txBody></p:sp>
    </p:spTree>
  </p:cSld>
</p:notes>

{
  "metadata": {
    "title": "Q1 Results",
    "author": "Marketing Team",
    "slides": 24
  },
  "blocks": [
    {
      "type": "section",
      "kind": "slide",
      "blocks": [
        {
          "type": "heading",
          "level": 1,
          "text": "Q1 Results"
        },
        {
          "type": "text",
          "text": "Revenue up 15% YoY"
        },
        {
          "type": "text",
          "text": "New markets: APAC, LATAM"
        },
        {
          "type": "text",
          "style": "speaker-notes",
          "text": "Mention APAC expansion
timeline: Q3 launch"
        }
      ]
    }
  ]
}

What Gets Extracted

Speaker Notes

PPTX stores speaker notes in separate notesSlide XML files, one per slide. These contain the presenter's talking points, data sources, and context that doesn't appear on the projected slide. AILANG Parse extracts notes as text blocks with style: "speaker-notes".

This is critical for AI applications: the notes often contain more actionable content than the slide bullet points. A training deck's slide says "Security Best Practices" — the notes contain the actual procedures and exceptions.

[
  {"type": "heading", "level": 1, "text": "Security Best Practices"},
  {"type": "text", "text": "Follow the principle of least privilege"},
  {"type": "text", "text": "Rotate credentials every 90 days"},
  {"type": "text", "style": "speaker-notes",
   "text": "Exception: service accounts in prod-critical path use 180-day rotation per SEC-2024-003. Mention the incident from January if anyone pushes back on rotation frequency."}
]

Tables in Slides

Presentation slides frequently contain data tables — quarterly results, comparison matrices, feature lists. These use the same a:tbl markup as DOCX tables, with gridSpan for horizontal merges and vMerge for vertical merges.

{
  "type": "table",
  "headers": [
    {"text": "Feature"},
    {"text": "Free"},
    {"text": "Pro"},
    {"text": "Business"}
  ],
  "rows": [
    [{"text": "API Requests"}, {"text": "1K/mo"}, {"text": "100K/mo"}, {"text": "500K/mo"}],
    [{"text": "File Size"}, {"text": "10 MB"}, {"text": "25 MB"}, {"text": "50 MB"}],
    [{"text": "Support"}, {"text": "Community"}, {"text": "Email"}, {"text": "Dedicated"}]
  ]
}

See Tables & Merged Cells for the full specification.

Use Cases

Try It

CLI

# Parse a PPTX file
ailang run --entry main --caps IO,FS,Env \
  docparse/main.ail presentation.pptx

# Convert PPTX to Markdown
./bin/docparse presentation.pptx --convert output.md

# Convert PPTX to HTML
./bin/docparse presentation.pptx --convert output.html

API

curl -X POST https://docparse.ailang.sunholo.com/api/v1/parse \
  -H "Content-Type: application/json" \
  -d '{"filepath":"sample_pptx_formatting","outputFormat":"markdown","apiKey":"YOUR_API_KEY"}'

Python SDK

from ailang_parse import DocParse

client = DocParse(api_key="YOUR_API_KEY")
result = client.parse("presentation.pptx", output_format="json")

# Iterate slides
for block in result.blocks:
    if block.kind == "slide":
        for child in block.blocks:
            if child.type == "heading":
                print(f"Slide: {child.text}")
            if child.style == "speaker-notes":
                print(f"  Notes: {child.text[:80]}...")

Parse in Browser    API Reference