i think this would probably just a matter of choosing the appropriate values for controlling the Granulator.
i think to compose rhythmically on a microlevel you could for example either use Demand rate Ugens to form microrhytmical gestures often called streams or clouds and sequence these clouds with patterns on a macro level or for example use recursive Phrasing which also follows the logic of micro and macro structure.
Here you find some other strategies for Buffer Granulation
But maybe someone else has a more straight forward approach which suits your needs.
Ive also found this discussion Rhythmically skipping through a recording, audio probing