ADMAntwerp/XAIstories — reverse-engineered prompt

Build me a Python tool that helps explain machine learning predictions by turning SHAP style feature explanations into simple narrative stories.

I want to be able to give it a trained classification model that has a predict_proba method, some input data, and a specific example to explain. The tool should calculate or use the explanation values, identify the most important features, then ask a large language model to write a clear story that explains why the model made that prediction in plain English.

Please include a working Jupyter notebook example that shows the full flow from loading data and a model, generating the explanation, and producing the final story. It should be easy to swap in another compatible model later. Also make it clear where API keys need to go for the language model part, and add simple setup instructions with requirements so I can run it locally.