Introducing: The Agentic RAG Chronicles (Blog Series)
Last year I chronicled several RAG Demos with the idea I would have a documented place to return to when I was looking for specify RAG features. Having those demos documented and all the reference links in one location really was useful for me - so I am going to do the same thing this year, but with Agentic RAG components.
Agentic RAG
Last year when I was learning RAG, that technique became known as Standard RAG or Semantic RAG. At its core, it’s a system that uses semantic search—understanding meaning rather than just matching keywords—to find relevant information, then feeds that information to an LLM to answer a user’s question. When you first build one of these systems, the results feel almost magical. But it doesn’t take long before you encounter questions the system can’t answer using semantic search alone.
For me, agentic RAG is a system of components with different retrieval systems, routers, verifiers, etc. The ultimate goal is to handle the 8 question complexity types (taken from a presentation I gave last yer, but defined in the Mintaka repo):
What I’m Looking For
This blog series I will be focusing on the various components needed for an Agentic RAG solution. I won’t be focusing solely on full application demos this time. To help these blogs be a better reference, I’ll be grouping the components (ie. router/intent classifier, retrieval systems, answer verifier, etc.)
Blogs in this series (to be updated as they get published):
Router/intent classifiers
- First Coming Soon
BTW: if you have any suggestions, please message me at @haleyjason on twitter/X.