@rjurney saw your showcase too and had that question of that pipeline. I have pipelines that essentially generate arbitrary data tables eg for concept expansion. And some sort of KG fits there perfectly, but at the same time the prompts/pipelines themselves are also elements in a table.
The prompts to BAML pipeline I’ve taken a pause at because the prompts work so well because of the context, and the output isn’t rich without the prompt’s programmatic/lived pretext