As a database person, I think most ORMs lack the ability to run complex queries. Sure, for a simple OLTP system, you might just need a few simple joins. But if you store valuable data (otherwise why store it), you will eventually want to ask questions about it. So this ORM positively surprised me, as you can still just use SQL and get a JS function generated from it. To bridge object-to-relational mapping, Oracle has…
You can definitely tell when an ORM was written by people who actually -like- databases rather than people who want to not have to think about them. perl's DBIx::Class ORM had a rule from day one of "if you can't get the exact same SQL you'd've written by hand, that's either a bug or a missing feature" and close to two decades on people who've moved to other languages still regularly tell me they miss that aesthetic.
Mongo just doesn't feel like that. Every example I have seen is simply "KV lookup". No discussion of things like range based queries, or how data relates to each other. Yes of course these things are possible, but it requires writing a imperative query. Seems like it encourages people to be lazy and duplicate data everywhere, which results in a poor data model.