Continuing my exploration of data stores for RAG (Retrieval Augmented Generation), I'm now digging into graph databases and GraphRAG.

Continuing my exploration of data stores for RAG (Retrieval Augmented Generation), I'm now digging into graph databases and GraphRAG.

I can conceptually see the advantages of a knowledge graph to power RAG vs. a pure vector data store.

The issue with vector database RAG is that you can sort of lose something in translation when all you do is come with K matching documents based on the prompt. Those documents *may* contain the right information to answer the question, but it's also possible that the best answer exists in the relationship between things not as discrete things in any of the documents.

I'm still very much a rookie on all this stuff, but it's fun to think about and play with.

This post was originally shared by on Linkedin.