A lot of the Sr. system architects I’ve talked to complained how time consuming/complex (lack of a better word) it is while managing both relational and graph databases. Every CRUD operation leads to more dusting/cleaning your latent factors/connections in your graph databases. There is ongoing research that the complexity arises due to the independent nature of your data across multiple databases, but the latent factors by definition are data-about-your-data. So the task is to make your latent factors dependent to their pertaining relational data, which enables easier house cleaning.