Amazon Bedrock & HealthLake automate healthcare claims with agentic AI pipeline

AWS has published a detailed walkthrough on building an agentic AI pipeline that automates healthcare claims processing from paper forms to FHIR resources. The solution uses three core AWS services: Amazon Bedrock Data Automation for intelligent document extraction from scanned CMS-1500 claim forms, Amazon Bedrock AgentCore to host a Strands agent that validates extracted data, and AWS HealthLake as the FHIR-compliant data store. When a PDF is uploaded to S3, an AWS Lambda function triggers Bedrock Data Automation to extract structured JSON output. That JSON is then passed to the agent, which searches HealthLake for matching patient, provider, and coverage records. The agent uses two tools – search_fhir_resources and create_fhir_claim – and includes a retry logic with prompt-based reattempts if initial searches fail. If all validations pass, the agent creates a standardized FHIR claim resource in HealthLake. Finally, the pipeline sends an SNS notification with either a technical summary for claims processors or a patient-friendly explanation. Amazon Bedrock Data Automation combines OCR, ML models, and generative AI to extract fields with confidence scores and bounding box data. Pre-built blueprints or custom configurations guide extraction, ensuring a predictable JSON output. AgentCore hosts the Strands agent, which uses the extracted data to look up existing records and build a FHIR representation. The agent also handles edge cases with a three-attempt validation strategy, focusing on high-confidence attributes to resolve missing references. This end-to-end workflow reduces manual processing time while maintaining accuracy through automated checks and intelligent retries. The entire pipeline includes deterministic supervisor logic in Lambda to ensure every document is processed or sent to a dead letter queue for exception handling.