MedVision + Explainability: Medical Image Analysis with Interpretable AI

HiperHealth's clinical pipeline currently lacks visual diagnostic capabilities. This project adds a modular MedVision skill channel, an external, installable module that brings medical image analysis into the existing pipeline with zero core modifications. The project delivers shared infrastructure (MONAI preprocessing, Grad-CAM explainability, structured Pydantic output models, confidence calibration) and two body-region skills: SkinAnalysisSkill (DINOv2 backbone, ISIC 2018 dataset) and a second body-region skill (Eye/Nail) to validate extensibility. Each skill automatically enriches the DiagnosticsSkill's differential diagnosis via the prompt_fragments mechanism, producing calibrated predictions with saliency maps for clinical transparency. The project includes Fitzpatrick fairness auditing and ≥90% test coverage.