Shawn Albert

Menu

Close

Agentic AI Clinical Documentation Pipeline

Serverless AI pipeline that transforms patient conversations into clinical documentation using AWS's AI services for automated medical transcription, SOAP note generation, and medical entity extraction.

Amazon BedrockAmazon Transcribe MedicalAmazon Comprehend MedicalAWS Step FunctionsApache IcebergAWS AthenaAWS S3ServerlessAWS SNSAWS ChatbotAWS Cloudwatch
AWS Cloud
Amazon
CloudWatch
Amazon Simple
Notification Service
AWS Chatbot
Slack Channel
CloudWatch Alarm to
notify of SNS failure
Pipeline Success or Failure Notification
Care Manager Phone Call 
Audio File (MP3)
AWS Step Functions workflow
Transcribed Audio
Amazon
Transcribe Medical
Processed Files
Clinical SOAP Notes
Amazon
Bedrock
Extracted Medical Entities
Amazon
Comprehend Medical
Append Outputs to
Apache Iceberg Table
Amazon
Athena
Landing S3 Bucket

Overview

Built an autonomous, serverless pipeline that transforms care manager-patient conversations into clinical documentation and billable insights. The system orchestrates multiple AWS AI services through Step Functions to create a fully automated documentation workflow with comprehensive monitoring and alerting.

Pipeline Flow

  1. Care manager calls are automatically captured and landed in S3
  2. Event-driven triggers initiate the serverless processing pipeline
  3. Amazon Transcribe Medical converts medical conversations to accurate text
  4. Amazon Bedrock generates comprehensive SOAP notes from transcriptions
  5. Amazon Comprehend Medical extracts medical entities
  6. Processed data is stored in Apache Iceberg format via Athena for downstream analysis
  7. Pipeline status monitoring and notifications:
    • Step Function completion triggers SNS notification
    • Success updates routed to team Slack channel via AWS Chatbot
    • Failure alerts trigger immediate team notification
    • CloudWatch alarms monitor notification delivery and email team members on failure

Technical Architecture

  • Implemented event-driven S3 triggers for automated audio file processing
  • Leveraged Amazon Transcribe Medical for accurate medical transcription
  • Utilized Amazon Bedrock for generating comprehensive SOAP notes
  • Integrated Amazon Comprehend Medical to extract:
    • ICD-10 diagnostic codes
    • Prescription information
    • Social determinants of health
    • SNOMED clinical terms
  • Stored processed data in queryable format enabling reporting and ML model development
  • Configured SNS topics and subscriptions for automated alerting
  • Implemented AWS Chatbot integration for streamlined incident response

Key Achievements

  • Built HIPAA-compliant data workflows implementing data encryption and role-based access controls with least-privilege principles
  • Designed scalable architecture handling parallel processing of multiple care manager calls
  • Implemented comprehensive error handling with automated notification system
  • Created real-time monitoring dashboard with Chatbot-enabled troubleshooting
  • Significantly reduced manual documentation overhead while improving billing accuracy
  • Enabled downstream healthcare analytics by storing JSON data in Apache Iceberg tables integrated with AWS Glue Data Catalog

Impact

The pipeline transforms the clinical documentation process from a manual task to an automated workflow, enabling healthcare providers to maximize direct patient care time while maintaining documentation accuracy and compliance standards. The robust monitoring and notification system ensures reliable operation with minimal manual oversight.