Implement information extraction solutions 5.2 Extract content from documents
You are ingesting reports full of charts and tables into a RAG index and want clean…
Study guide skill: Produce clean, grounded representations to use with agents and RAG by using Content Understanding
Question B1-050
You are ingesting reports full of charts and tables into a RAG index and want clean Markdown that keeps the layout and includes descriptions of the figures, with minimal custom code. What should you use?
Select one answer.
Answer: A
Content Understanding provides prebuilt analyzers optimized for RAG that extract Markdown for document ingestion. It keeps structure such as tables and sections and adds figure descriptions so charts and diagrams become searchable. The other services translate, transcribe or generate content rather than extract it.
Deep dive
How it works
Azure Content Understanding uses generative AI to process documents, images, video and audio into a format you define. For search and RAG, it can ingest content of any modality into an index, with extensive support for figure description and analysis, and it offers several prebuilt analyzers tuned for RAG scenarios.
The RAG analyzers are optimized for retrieval-augmented generation, with semantic analysis and Markdown extraction for document ingestion, search and knowledge bases. For documents, prebuilt-documentSearch:
- extracts paragraphs, tables and figures;
- gives detailed figure descriptions that explain images, charts and diagrams, with charts output as chart.js syntax and diagrams as mermaid.js syntax;
- captures handwritten annotations and generates a one-paragraph summary;
- produces chunked output ready for embedding, keeping document structure across chunk boundaries with fixed-size or layout-aware semantic chunking.
That's reports full of charts and tables turned into clean, searchable Markdown with minimal custom code. You can also customize prebuilt analyzers to extract extra fields.
The other options, and when you'd use them
- Azure AI Speech batch transcription: Batch transcription transcribes large amounts of stored audio. It's for audio, not documents.
- The Azure OpenAI image generation API: Image generation models create images from text prompts and optional images. They produce images rather than extract them.
- Azure Translator document translation: Document translation translates documents while keeping their structure and format. It changes the language, not the representation for indexing.
Exam tip
"Markdown plus figure descriptions for a RAG index" means Content Understanding's RAG analyzers, such as prebuilt-documentSearch. If you only need layout, prebuilt-layout extracts words, figures, paragraphs and tables and identifies document structure.
References
Last reviewed: