Retrieval should also do a full-text search across chunks and metadata, and those results should be combined/reranked together with the vector search retrieved results.

THIS CAN BE DONE WITH QDRANT, BUT REQUIRES CHANGING THE COLLECTION STRUCTURE:

• https://chatgpt.com/share/68aed5fd-1734-8003-b304-5fbc37951f36
• https://qdrant.tech/documentation/concepts/hybrid-queries/#hybrid-search
• ( https://chatgpt.com/share/688a6061-57ec-8003-a91d-fa3d2d4e83f4 )

See also:

• https://www.innoq.com/en/blog/2024/12/rag-hybrid-search/
• https://techcommunity.microsoft.com/blog/azuredevcommunityblog/doing-rag-vector-search-is-not-enough/4161073