Most RAG tutorials teach you one pattern and leave you to figure out the rest. The reality is that a Basic RAG pipeline that performs brilliantly for a documentation Q&A bot will fail in ways that are embarrassingly hard to debug when you add multi-turn conversations, stale retrieval, or complex multi-step tasks.

This session introduces a practical, opinionated framework for matching RAG architectures to your actual requirements, built around three tiers: Good, Better, and Best.

This session is geared toward Data architects, engineers, developers, and anyone interested in RAG with OpenSearch.

Throughout our session we will start with a foundational use-case and dataset and show the difference between:

“Good” (Simple RAG)

“Better” (Hybrid RAG)

“Best” (Agentic RAG with Memory)


We’ll discuss the importance and differences of each variation, and not just tell you, SHOW you (Live demos are great, right?!)

By the end of this session, you'll know which RAG architecture fits your use case, what the real tradeoffs are between each approach, and how to implement all three with OpenSearch.