Is this really RAG?

[pors]

I had a quick look at the code, but this looks more like a Retrieval-Generator architecture than RAG. In RAG the retriever is included in the model, in this app, the model is a chat model using a vector store for retrieval.

[athrael-soju]

@pors If you look at the video by Roie and Zack they go over this here: https://youtu.be/pPbTzgUV-UA?si=5AGavnBcP5mQi3u6&t=333

This diagram pretty much explains it:

In this example, RAG is essentially the results from pinecone after enhancing the mesages array with a simple prompt (src/app/api/chat/route.ts). No use langchain, but that's not necessarily bad, because it keeps the code more readable.

RAG can infinitely more complex, but for the sake of the example, it's more than enough.

[pors]

Thanks for the response! I see now that RAG is the term everyone uses for this architecture. In the "past" this term was coined by Meta for something similar but as part of the model itself. Not anymore, though :)