JensBender/loan-default-prediction ? reverse-engineered prompt

Build me an end to end loan default prediction app that takes customer application data and tells me how likely a loan is to default.

I want the full workflow included, from cleaning and preparing the data, handling missing values and outliers, creating a few useful features, training and comparing a handful of models, then picking the best one and using a better threshold for catching defaults. Please make it show some basic insights too, like which features matter most and a simple way to see example predictions.

Also set it up as a working web app with a simple front end and an API behind it, and package everything so it runs cleanly in Docker. If it helps, look up current docs online for any setup details. Please include tests for the main pieces so I can trust the pipeline and the app actually work.