Directed acyclic graphs to take care of any data and resource inter-dependencies and just focus on having a stable output.

Basically similar to e.g. dagster or other high-fallutin' cloud offerings, and somewhat similar to targets for R.
Of primary interest here is the cache=True setting for a pipeline which ensures its underlying functions only get recomputed when they actually change (see the ipynb example Cache section).

Seems super nifty just because it's super lightweight and easy to get started - but is fairly new as of this moment.