Adding Vector Search and Embedded Python to a FHIR Patient Snapshot Agent

Details

The agent retrieves structured FHIR resources (Patient, Condition, MedicationRequest, etc.) for a selected patient and normalizes them into a compact context. This data is used to generate either deterministic or LLM-generated role-specific summaries. The two main additions are: first, 'Source Context Vector Search,' which allows users to quickly identify and display the most relevant sections within the normalized patient context when asking questions about specific information (e.g., medications, allergies). Second, 'Embedded Python Artifacts.' This demonstrates embedding Python logic directly into InterSystems IRIS class methods. Examples include extracting numeric Observation values or calculating cosine similarity, showing a review process suitable for the IRIS environment. Furthermore, the tool now includes the ability to transform FHIR Observation resources into visual charts (for vital signs, lab results, etc.), providing compact visual summaries alongside the text snapshot. These enhancements aim to increase its practical applicability in clinical settings.