appreciated . There doesn't seem to be enough time in the day to "learn it all" ha. My thoughts: - graphRAG has potential but it's still too costly & labor intensive to get good results. It really depends on high quality entity creation. Also speed of retrieval seems a minus for most use-cases. - vectorRAG has still my preference although it's currently limited by the quality of the data we put in and the labor it requires to inject global context awareness (trough meta-data or some kind of Parent Child system). - Another downside of GraphRag/Vision Based RAG is that data is indexed in such a way that it becomes difficult to build applications. With generalized (read: parsed data to text) RAG it is relatively simple to build new use-cases with due the nature of text being processible vs. images/edges.
![[Part 1] ทารกน้อยถูกทิ้ง👶 หญิงชราเก็บขวดขายเลี้ยงดูเขาจนโต #โลกการละคร](http://i.ytimg.com/vi/nPTDvFCExks/mqdefault.jpg)
appreciated . There doesn't seem to be enough time in the day to "learn it all" ha. My thoughts:
- graphRAG has potential but it's still too costly & labor intensive to get good results. It really depends on high quality entity creation. Also speed of retrieval seems a minus for most use-cases.
- vectorRAG has still my preference although it's currently limited by the quality of the data we put in and the labor it requires to inject global context awareness (trough meta-data or some kind of Parent Child system).
- Another downside of GraphRag/Vision Based RAG is that data is indexed in such a way that it becomes difficult to build applications. With generalized (read: parsed data to text) RAG it is relatively simple to build new use-cases with due the nature of text being processible vs. images/edges.
I am trying to do this with data that I have in JSON format (so I can keep meta data). Wondering if that's possible
bangbangde