Atlantic Health cuts legal query response time 42% with Amazon Bedrock metadata filtering

Atlantic Health, based in Morristown, New Jersey, reports a 42% reduction in average query response time due to reduced search space after deploying metadata-filtered retrieval-augmented generation (RAG) for its legal department. The AWS blog post, co-written by Atlantic Health staff, also lists an 87% improvement in returning relevant documents for hospital-specific queries, 94% of legal department users finding the correct information on the first query, and a 35% reduction in OpenSearch compute costs from smaller result sets. These are self-reported figures. Working across multiple hospital systems, the legal team found that conventional keyword search could not reliably distinguish between closely related legal terms tied to different hospitals or time periods, leading to inefficient retrieval and exposure to compliance risk. Each hospital operates under distinct regulatory frameworks, with documents effective for specific periods. Metadata filtering also reduced false positives where documents from the wrong hospital or time period appeared in results. Legal documents from different hospitals are uploaded to an Amazon SFTP file server backed by an Amazon S3 bucket. An AWS Lambda function ingests them into Amazon Bedrock Knowledge Bases, and Amazon Bedrock Data Automation extracts entity metadata such as hospital name and effective year, which is stored in Amazon OpenSearch Service alongside the vector embeddings. Users ask questions in a chat application and can pass hospital or effective-year filters as parameters in the invoke agent call. Amazon Bedrock Agents orchestrate retrieval, using Amazon Bedrock foundation models to understand intent, querying OpenSearch with combined semantic search and metadata filtering, and keeping the user's conversation history in agent memory. The generated response, with source citations and metadata context, is returned through an application hosted on Amazon CloudFront and Amazon ECS with AWS Fargate. Amazon Bedrock Guardrails evaluates each response and applies safety, privacy and responsible AI controls based on Atlantic Health guidelines. The authors are Vishal Grover and Kunal Pitale of Atlantic Health and Karan Pande of AWS.