An algorithm that adapts more personally to the individual rather than a universal standard

By the time someone gets through about 20 workouts, it would be quite easy to see the convexity of the training load function. Most standard training load charts I see that predict 1RM are quite linear. In my experience, this is not accurate enough and changes based on experience. Also, some people simply have higher endurance which allows them to get a couple more reps at lower loads, but doesn't translate to the typical higher 1RM. In this case, in your app, the person will hit a strength phase having done bench @225 for 8 reps and suddenly it jumps to 250 expecting 5 reps, but that person just has high endurance but lower relative strength at low reps. I've easily plotted this function for myself, so I know a reasonable data scientist can feed an adaptive convexity argument into the AI function.