Copyright of aggregated and curated LLM output has not been decided upon, but other curation precedents suggest it is going to be copyrightable, similar to how curated collections of facts are copyrightable, even though facts themselves are not.

Why do you think it's not copyrightable? It absolutely is under current US law. OP copied the interfaces, which is perfectly legal, and then defined their own implementations. Even if they are working with an LLM, that process of interactive prompting establishes authorship.