Improved experimentation cabalility.

[deatheragetr]

It would be nice to be able to use a consistent corpus of raw data (like what gets stored in Mongo) and use different models and techniques to easily generate vector embeddings on that data that could all be store in Postgres together, and selectively queried.

Some ideas.

• Separate tables or separate columns or even separate databases for the different embeddings, each legibly-labeled.
• The ability to choose which model's embbeddings are being queried from the API. Could be an env var that's set at start up of the backend API, or an optional parameter passed to the API.