Docling for IBM watsonx, available on AWS Marketplace, transforms unstructured enterprise documents into structured, AI-ready data for generative AI applications on AWS. Generative AI applications depend on high-quality, structured data, but unstructured documents such as PDFs, technical manuals, financial reports, contracts, and presentations contain valuable business knowledge designed for people to read, not for AI systems. Organizations adopt architectures like retrieval-augmented generation (RAG), enterprise search, and AI agents to connect foundation models to enterprise content, retrieving information from existing documents instead of relying on pre-trained knowledge. Their effectiveness depends on the quality of retrieved information; poor document preparation leads to incomplete retrieval, inaccurate responses, and reduced trust. Most enterprise AI architectures store documents in Amazon Simple Storage Service (Amazon S3), with a search engine or vector database indexing content for large language models. Extracting text often strips context like document hierarchy, tables, reading order, images, captions, and page layouts, causing retrieval pipelines to return incomplete context. Docling addresses this by preserving the semantic structure and layout of original documents, understanding and retaining context instead of treating every document as plain text. The post explores a reference architecture and integration patterns to improve retrieval quality for enterprise search, AI agents, and generative AI applications on AWS.