aws-samples/layout-aware-document-processing-and-retrieval-augmented-generation

Leverages Amazon Textract's layout detection to preserve semantic structure (titles, headers, tables, lists) and enriches extracted text with XML tags, enabling intelligent chunking that maintains context hierarchies. Implements specialized chunking strategies—tables preserve column headers per chunk, lists include list titles, and text sections carry parent headers—creating a child-section-chapter metadata hierarchy stored in OpenSearch. Integrates with Amazon Bedrock, SageMaker JumpStart embeddings, and the Textractor library to enable flexible RAG retrieval using hybrid search and hierarchical context selection.

115 stars.