Develop a framework for medical agents that persists a 'case-memory' database to allow models to learn from past diagnostic errors. This enables persistent improvement in specialized agent workflows without full fine-tuning.