In case folks are interested: Looking for the best way to add ML model predictions to data we have in Snowflake and still keep as much as possible where dbt can "see" it and manage the tables and stuff.

Original plan was to regularly have Prefect run inference jobs on a target table, export a flat file to a Snowflake stage, then add that as a Data Source for dbt

Buuut then I saw this article, and wondered if it'd make sense to do this (except with Prefect instead of SageMaker): https://community.snowflake.com/s/article/How-To-Use-Amazon-SageMaker-model-with-data-in-Snowflake-directly-in-SQL

Which has the advantage of running automatically every time dbt rebuilds the tables

Although, I guess a downside vs "just run bulk jobs" is that it'd probably be less efficient compute-wise...