-
Notifications
You must be signed in to change notification settings - Fork 127
Description
Hey team,
current RAG availability in Kaiban is very limited. Its doing mostly basic RAG aka Naive RAG. But we definitely need more advanced strategies available in terms of retrieval, chunking etc. Current most of the frameworks are in Python like microsoft graphRAG, lightRag, nanoRAG etc. Also chunking techniques like https://github.com/GPTim/chonkie etc. These are just for reference.
There is another TS framework known as mastra which does this quite good. Sharing a link here for reference https://mastra.ai/docs/reference/rag/graph-rag. I believe to make Kaiban as the best JS framework we need to make it more robust because RAG is such use case which cannot be denied and more & more sophisticated techniques are being built to make it more accurate and robust.