Query claims in natural language with Amazon Bedrock Knowledge Bases
aws ·
This article presents a technical guide for building a claims assistant using Amazon Bedrock Knowledge Bases to query unstructured insurance documents. The solution addresses the challenge of scattered claim data by employing Retrieval Augmented Generation (RAG) to provide cited, natural-language answers. Key components include ingesting PDF and Word files from Amazon S3 into a managed knowledge base with metadata sidecars for filtering. The system utilizes the AgenticRetrieveStream API to handle multi-part questions and conversation history through iterative sub-queries. Security and accuracy are ensured via metadata filters for access control and Amazon Bedrock Guardrails for contextual grounding. Performance metrics on synthetic data show high retrieval recall and consistent citation generation across various claim scenarios.
aws