Build a Claims Assistant Using Amazon Bedrock Knowledge Bases

Amazon Bedrock Knowledge Bases enables a conversational claims assistant that answers natural-language questions with citations drawn from PDF adjuster reports, Word correspondence, and text notes stored in Amazon S3. The solution uses the AgenticRetrieveStream API to break multi-part questions into sub-queries, run iterative retrieval passes, and stream answers with source citations. A contextual grounding guardrail blocks responses unsupported by retrieved records. Metadata filters let users scope searches by claim ID or type. The walkthrough covers document ingestion, plain-language querying, multi-turn follow-up questions, and guardrail configuration using synthetic claim records.