RAPTOR in a production environment

[atirathk]

I was thinking about how something like this would be deployed in a production environment. If there's a lot of text data to go over, it obviously would not be a wise idea to store all the data in one data base. I'm thinking the nodes could be stored in a graph data base, and each node could have a foreign key to a vector data base where the embedding for that node's text is stored. I'm thinking sharding could be used for horizonal scaling as well. Maybe the nodes could be partitioned by the subtrees they belong to, or they could be partitioned randomly.

Has anyone else been thinking about how the data could be stored?