Machine Learning
Amazon Textract
Amazon Textract extracts text, forms, tables, queries, signatures, and layout from scanned documents and images. It produces structured blocks that can preserve document relationships before chunking or prompting.
Key points
- Synchronous APIs fit small supported documents while asynchronous jobs handle multipage files
- Geometry and relationship fields preserve page structure
- Confidence scores enable review thresholds for critical fields
When to use it
- Extract tables and headings for layout-aware RAG chunks
- Read invoice fields before a generative reconciliation step
Exam tips
- Choose Textract for document structure and Rekognition for general image analysis
- Do not flatten blocks blindly; preserve page, table, and key-value relationships and review low-confidence fields