I put together four self contained notebooks exploring different ways to make a RAG pipeline more agentic.
Instead of retrieving once and immediately generating an answer this methods show patterns where the system can perform the following
* decide whether retrieval is needed
* where to retrieve from
* evaluate retrieved context
* correct or retry when the context isn't sufficient
* defer to a human
Each notebook focuses on a different pattern and is intended to be runnable and easy to experiment with.
I put together four self contained notebooks exploring different ways to make a RAG pipeline more agentic.
Instead of retrieving once and immediately generating an answer this methods show patterns where the system can perform the following
* decide whether retrieval is needed * where to retrieve from * evaluate retrieved context * correct or retry when the context isn't sufficient * defer to a human
Each notebook focuses on a different pattern and is intended to be runnable and easy to experiment with.
The notebooks are free to explore on GitHub https://github.com/ChandulaSenevirathna/Agentic_RAG
I also put together a more complete version with all four implementations and explanations here https://chandula7.gumroad.com/l/Advanced_RAG_LangGraph_Patte...
Would be interested in feedback on which Agentic RAG pattern people are finding most useful in real projects.