I have a clarification question regarding the search component of the MM-RAG pipeline. Since effective retrieval is crucial to RAG performance, I’m wondering:
Is it permitted to re-structure or augment information from the indexes (e.g., with additional metadata, alternate text representations, or even domain-specific information, naturally processed from the original data) to have a user defined search mechanism, or are they expected to be used as-is/intact through the CRAG Search API?
Understanding the flexibility we have here would really help us design better multi-modal retrieval strategies. Appreciate any clarification!