Provides an abstraction layer for accessing patient data from multiple sources (Azure Blob Storage, FHIR, Microsoft Fabric) through a common interface. Demonstrates clean separation of concerns by encapsulating source-specific logic behind interchangeable implementations, enabling easy extension to new data sources without modifying core agent logic.

A reusable HTTP client wrapper that adds standardized logging and error handling to outgoing requests. Self-contained with minimal dependencies (only httpx and logging), it can be extracted to any project requiring reliable HTTP communication with observability features.

Defines an abstract base class for evaluation metrics that enables consistent implementation of both reference-based (e.g., ROUGE) and LLM-as-judge metrics. Demonstrates extensible design where new metrics can be added by inheriting from the base class and implementing a single evaluate method, promoting reuse across different evaluation scenarios.