Generative System with built-in constraints

Hello,

I have a fairly complicated idea that I've been attempting to realise through generative systems but cannot get anything to work. Bare with me, I'll try explain as best I can!

The system follows the decimal series of pi and generates a midi output (say to be played on the diskclavier) that "flows" (sounds cohesive, from one 'segment' (say, bar) to another whether that be melodically, harmonically, etc.). To do this, I have paired each digit of the decimal series with the immediate succeeding digit to form these 'segments.' These two-digit combinations determine how many notes are played in each segment, and how many times it is played.

I.e., 3.14159... {14} - 1 note, 4 times; {41} - 4 notes, 1 time; {15} - 1 note, 5 times.

I need the system to do two things:

1. creative a cohesive "flow" from segment to segment, so it doesn't sound like a random jumble of notes

2. remember the segments that have come before (so that it can create this cohesion) but to also recognise when the two-digit combination appears again, so that it immediately "plugs-in" what had been already written (but importantly, not use that plugged-in bar to then influence the cohesion of the segments that immediately succeed).

The idea of the work is that as the decimal series is played through, the "flow" will begin to falter as more and more two-digit combinations re-appear.

Any tips? Or perhaps an example of a similar concept that I can try and alter to produce this idea? Any insight would be great.