Today, AWS announces the general availability of multimodal retrieval in Bedrock Knowledge Bases. Amazon Bedrock Knowledge Bases offers managed, end-to-end Retrieval Augmented Generation (RAG) workflows to create accurate, low latency, and custom Generative AI applications by incorporating contextual information from your company’s data sources. Supporting multimodal retrieval in Knowlesdge Bases enables developers to build AI-powered search and question-answering applications that work across text, images, audio, and video files. For example, a user could ask their assistant “show me Q1 projections for Amazon Bedrock” and Bedrock Knowledge Bases will retrieve relevant text from documents, graphs, video snippets, and audio related to revenue projections for Bedrock, allowing the assistant to generate richer and more complete answers for the end user. Previously, customers could only search through text documents and images. Now they can unlock insights from all their enterprise data formats through one unified, fully managed workflow.
Organizations struggle to extract insights from their growing multimedia data—videos, audio recordings, images, and documents— because building AI applications that can search across these different modalities is complex. As a result, valuable information trapped in terabytes of meeting recordings, training videos, and visual documentation remains inaccessible, preventing organizations from making data-driven decisions quickly and accurately. With multimodal retrieval for Knowledge Bases, developers can ingest multimodal content with full control of the parsing, chunking, embedding (e.g. Amazon Nova multimodal), and vector storage options. From there, they can then send a text query or an image as input and get relevant text, image, audio, and video segments back in order to generate a response in their generative AI applications using their choice of LLM.
For more information about creating multimodal Knowledge Bases in Bedrock, please refer to the documentation. Region availability is dependent on the features selected for multimodal support, please refer to the documentation for details.
Categories: general:products/amazon-bedrock,marketing:marchitecture/artificial-intelligence
Source: Amazon Web Services
