wtf is loan-default-mlops?

loan-default-mlops is a portfolio project that demonstrates how to manage a machine learning model after it has been trained, not just how to train one. It is built around a synthetic dataset of 8,000 small business loan applications modeled on Nigerian fintech lending, including details like mobile money transaction patterns, airtime top-up frequency, and utility payment consistency, alongside more traditional loan information such as revenue, loan amount, and prior default history. The project trains four different types of models, logistic regression, random forest, XGBoost, and LightGBM, and tracks every training run using a tool called MLflow, which records each model's settings, performance metrics, and supporting charts like confusion matrices. The best performing model is automatically registered into a model registry with a staging status, and can later be promoted to a production status through a dashboard. A separate script simulates new incoming loan applications and checks whether their statistical distribution has drifted away from the original training data, which is a common early warning sign that a deployed model is starting to become less accurate over time. Once a model is promoted, it can be served as a web API that accepts a loan application's details and returns a prediction, and there is also a dashboard for browsing the model registry, comparing a currently deployed champion model against a challenger, manually promoting models, and checking drift status. The project includes example commands for starting each of these pieces: the tracking interface, the training script, the drift check, the dashboard, and the prediction API. Setup only requires installing the Python packages listed in the requirements file. The README also notes that for a real deployment beyond a portfolio demo, the tracking data should be stored somewhere persistent rather than in local files, and includes talking points meant to help explain the reasoning behind the project's design choices in a job interview setting.