eCR Semantic Interoperability
A pipeline that pulls structured clinical data out of electronic case reports for public health agencies, replacing a manual review process that couldn't keep pace with outbreak surveillance.
Public health agencies get flooded with electronic case reports (eCRs) where the clinically important details are buried in unstructured, free-form text and XML. Reading through all of it by hand doesn't scale, especially when speed matters for outbreak detection. This pipeline handles the XML preprocessing, resolves cross-references, and uses embeddings plus natural language inference to pull out and classify the relevant clinical information automatically, with JSON schema configs so it adapts to different report formats.
Worked on this with Gus Flusser and Ryan Gertz. Built on Amazon Bedrock with Titan Embeddings V2 and Claude Haiku, deployable on EC2, Lambda, or on-prem.