ZJU4HealthCare/HealthGPT ? reverse-engineered prompt

Build me a medical AI assistant that can read and reason over medical text and images, then answer questions clearly and generate useful medical style responses. I want it to feel like a unified multimodal model for healthcare, so it should handle things like radiology images, pathology images, scans, and plain medical text in one place, and support both understanding and generation.

Please make the project runnable end to end, with training and inference workflows, example prompts, and a simple way to try the model on sample inputs. If there are separate versions for a stronger understanding model and a more general comprehension plus generation model, set that up too. Keep the setup practical and document how to load models, prepare data, and run a demo. If anything depends on current model or library docs, look them up online and use the latest working approach.