Nothing about a deployed model tells you its inputs have changed. Accuracy needs labels, and labels arrive weeks late or never. What you have on the day is the feature itself, and the distribution it had at training time.
Write population_stability_index(expected, actual, n_bins=10).
expected is the training distribution, actual is what production is sending now. Cut expected into n_bins quantile bins, so each bin holds roughly the same number of training rows. Then bin both samples with those same edges. For each bin:
$$PSI = \sum_{b} (a_b - e_b) \, \ln \frac{a_b}{e_b}$$
where e_b and a_b are the fraction of each sample in bin b. The industry reading: below 0.1 is stable, 0.1–0.25 warrants a look, above 0.25 is a real shift.
Build the architecture on a canvas: place the components, configure them, connect them into a data flow, and write a short reason for each one. The AI reviewer grades your design against a rubric written specifically for this problem.
Minimum 5 components · needs a wide desktop screen