A new framework for medical text generation has been developed that improves response quality by using a severity-aware, multi-model approach. This method outperforms existing models by effectively adapting to varying case complexities, achieving notable accuracy improvements in medical responses.