Ofc I reserve the right to be wrong, just wanted to share my subjective experience, that there can be tradeoffs and there probably aren't any silver bullets.
For the most part, I think that you should put any mass/batch processing in the DB (just comment/version/test/deploy your code like you would on the back end, as best as you can with the tools available to you) and don't sweat too much about handling the CRUD operations in your back end, through whatever ORM you use or don't use (regular queries are also fine, as long as parametrized to prevent injection).
For complex schemas, a nice approach I've found is making one DB view per table/list/section of your front end, so you only need 1 DB call to load a particular component, otherwise the N+1 risk gets far greater ("Oh hey, I got this list of orders, but each other needs a delivery status, so I'll just iterate over those and fetch them for each item, whoops, the DB is spammed with requests.").
For the most part, I think that you should put any mass/batch processing in the DB (just comment/version/test/deploy your code like you would on the back end, as best as you can with the tools available to you) and don't sweat too much about handling the CRUD operations in your back end, through whatever ORM you use or don't use (regular queries are also fine, as long as parametrized to prevent injection).
For complex schemas, a nice approach I've found is making one DB view per table/list/section of your front end, so you only need 1 DB call to load a particular component, otherwise the N+1 risk gets far greater ("Oh hey, I got this list of orders, but each other needs a delivery status, so I'll just iterate over those and fetch them for each item, whoops, the DB is spammed with requests.").
Good luck!