Wiring the model into the app is the easy part now; I can do that in an afternoon. But once real users hit it, things break in ways I never saw coming, weird answers, random failures, stuff I can't reproduce. Getting it to just work, reliably, day after day, still beats me every time.
More people experience this than they realize.
You can plug in the model, but making it run reliably for real users still defeats you.
If this sounds familiar, the Library can help you find the bigger picture.