Implement information extraction solutions 5.1 Build retrieval and grounding pipelines
An existing Azure AI Search skillset that is already in production chunks long…
Study guide skill: Implement enrichment by using custom or built-in skills for text, images, and layout
Question B1-048
An existing Azure AI Search skillset that is already in production chunks long, structured PDFs by their headings and sections before vectorizing them. Which built-in skill in that existing pipeline produces a Markdown representation of the document's structure for this purpose?
Select one answer.
Answer: C
The Document Layout skill calls the Document Intelligence layout model to detect a document's structure and output it as Markdown or text, which enables chunking by document layout (including settings such as markdownHeaderDepth). It remains supported for existing pipelines; for new skillsets, the current guidance is to use the Azure Content Understanding skill, which combines content extraction and chunking. The other skills extract phrases, languages or entities, not structure.
Deep dive
How it works
The Document Layout skill uses the Azure Document Intelligence layout model to analyze a document, detect its structure and produce a representation in Markdown (the default) or text. That Markdown output is what lets an existing pipeline chunk long, structured PDFs by headings and sections before vectorizing them.
Settings worth knowing:
- outputFormat:
markdown(default) ortext. - markdownHeaderDepth: the deepest heading level to treat as its own section, from h1 to h6 (the default). For example, with h3, deeper sections such as h4 are rolled into h3.
- extractionOptions: can add images and location metadata, which helps RAG and multimodal search keep images close to related content.
Limits: the skill isn't suited to large documents that need more than five minutes of layout processing; it times out, but charges still apply. It needs a billable Foundry resource beyond 20 documents per indexer per day.
The Document Layout skill remains supported for existing pipelines. For new skillsets, the current guidance is the Azure Content Understanding skill, which combines content extraction and chunking and supports semantic chunking, AI-generated image descriptions and cross-page tables.
The other options, and when you'd use them
- Entity Recognition skill: Extracts entities such as people, organizations, URLs and phone numbers across 14 categories.
- Key Phrase Extraction skill: Returns key phrases for each record, which helps you spot the main talking points quickly.
- Language Detection skill: Reports a language code and score for each document, which is often passed to other skills such as Text Split.
Exam tip
"Markdown by headings for chunking" means Document Layout in an existing pipeline. For a brand-new multimodal pipeline, the recommended choice is the Content Understanding skill.
References
Microsoft Learn references
Last reviewed: