Ask HN: What is the state of art for doing RAG in Enterprise LLM agents?

[ppipada]

I am trying to understand and build RAG in few of the LLM Agents I am working on.

I have noted a small collection of things that are currently being done in the field at: https://pankajpipada.com/posts/2024-12-17-rag-resources/

From all the reading and experimenting it looks like Graph + vectors + optional pagerank for certain usecases gives best results.

Question:

For enterprise unstructured data (docs/logs/scripts/etc), what is the current state of art technology wrt establishing a knowledge base to be used in either search or context population for some code generation usecases?

Thanks in advance

[divineer]

Building and Managing RAG applications requires a new learning curve. I did work on Simple to use RAG APIs that you can use without worrying too much about the complexities. checkout https://docs.wetrocloud.com

[softwaredoug]

Don't underrate just having your agent understand how your traditional search engine works and generating well-understood keyword matches + filters. Agents are good at generating structured queries into specific fields, etc (this taxonomy category, a title that looks like this, etc)

https://softwaredoug.com/blog/2024/09/11/generativeai-augmen...