sarvanithin/Medguard

AI Summary: This issue proposes expanding the `drug_interactions.csv` file by incorporating over 500 high-risk drug pairs from DrugBank's open data. The task involves downloading, filtering, normalizing, and mapping the data to the existing CSV format, ensuring deduplication and adherence to specific acceptance criteria. This is a data-focused contribution with no code changes required.

AI Summary: This issue proposes enhancing the existing English-only PHI detection engine to support Spanish, French, and German patient data. It involves creating locale-specific regular expressions for identifying various identifiers like DNI, INSEE, and KVNR, and integrating them into a new `LocalePHIEngine` that can be configured with a specific locale.

AI Summary: This issue proposes integrating the Unified Medical Language System (UMLS) to enhance hallucination detection for medical terms. The current SNOMED-based system has limited coverage, and UMLS offers a vast repository of medical concepts. The implementation involves creating a new `UMLSClient` class, caching results, and making it a drop-in replacement for the existing SNOMED client, with a fallback mechanism if a UMLS API key is not provided.

AI Summary: This issue proposes to enhance the existing fact verification system by replacing keyword matching with an LLM agent. The agent will reason over PubMed search results to determine if a claim is supported, contradicted, or inconclusive, providing a more nuanced verdict. The new implementation will be opt-in and fall back to the keyword-based verifier when an LLM is not configured.