TABLE OF CONTENTS

• Purpose
• Development Process
• Understanding the Technology
• AI Model - Claude by Anthropic
• Nightly Batch Processing
• Next Steps/Future Development

Purpose

• How bhworks leverages AI to gain quick insights into documentation.

Development Process

• The infrastructure was built for a secure, HIPAA, DTMB, NIST 800-53 RMF compliant environment, approved by the state of Michigan
• De-identified case notes are used for testing in a fully compliant environment
• Human evaluation of Case Notes Summaries improves quality and accuracy
• User feedback is gathered to further improve implementation

Understanding the Technology

• We use an AWS-managed AI service
• This service provides access to models from leading AI companies (including: Google, Meta, and Anthropic)
• Data from bhworks is not shared with the model providers; it stays entirely within mdlogix/bhworks existing secure cloud
• The AI service is HIPAA and ClearDATA compliant

AI Model - Claude by Anthropic

• Large Language Model (LLM)
  • AI system trained on vast amounts of text to understand and generate human language
• State-of-the-art and very ethically conscious

Nightly Batch Processing

• Case Notes Summaries generated per enrollment
• Summaries are updated every night shortly after midnight
• Includes: Notes created/updated the previous day
• Excludes:
  • Future-dated notes
  • Single short notes (<50 words)
  • Notes older than 1 year
• When Case Notes Summarization is turned on for an organization, it will backfill summaries the following night

Next Steps/Future Development

• Real-time inference with flexible date range
• Method for saving summaries
• Improved implementation with user feedback