Lots of components including RAG and safety shield feature has changed since the update to 0,3 Llama stack repository: https://github.com/redhat-ai-dev/llama-stack

This task is to investigate whether the same approach taken in the Spike feature is still viable and the approach taken in the repository (https://github.com/JslYoon/rhdh-plugins/tree/JslYoon-ai-notebooks) is still optimal.

If features that was not available in Llama stack 0.2.x that could enhance AI Notebooks is available (i.e., metadata filtering, custom shield, document deletion), update the proposal accordingly.  

Acceptance criteria:

• Ensure RAG and safety shield is stable. Ensure there aren’t any additional changes to be made in the base image.
• Ensure the approach made for the spike issue is still feasible.
• If metadata filtering is enabled, revise the architecture to use it.
• Investigate whether document upsert/delete is possible
• Revise the AI Notebooks architecture if needed.