Llama-index is a super useful and performant tool for creating the vector store within iris, however the current integration between iris and llama-index doesn't support metadata (a capability of newer versions of llama-index). This would be a very helpful feature for many applications, for example where a hybrid search of documents is required (e.g. filtering by date or author of a document before doing a vector search). Currently, the workaround would be to build a normal SQL database and populate the embeddings, metadata, text chunks etc. yourself, however being able to do this directly with llama-index would be highly convenient.
Thank you for submitting the idea. The status has been changed to "Community Opportunity".
Stay tuned!
Somesh Mehra, thank you for your idea.
There is a comment on your idea from InterSystems Product Manager. Please answer it to help your idea to be promoted.
A similar suggestion was filed in the GitHub repo: https://github.com/caretdev/llama-iris/issues/1
We intend to look into adding this ourselves, but given this is a public repository, would also accept pull requests