We Built Our RAG Twice, So You Don’t Have To

This session presents a side by side comparison of two production RAG implementations solving different problems. One uses classic vector search for internal troubleshooting documentation. The other employs hierarchical LLM routing for structured product catalogs.

The contrast reveals critical questions: Is your content naturally unstructured or does it have implicit hierarchy? Do users ask "how to" questions or "which one" questions? Is semantic similarity what you actually need, or do you need categorical precision and exact matching?

You'll discover both architectures, understand why one retrieval strategy can't serve all content types, and leave with a decision framework for your own systems. Because sometimes the "advanced" architecture is overengineering, and the "simple" one is naive.

Youssef MouhimCIH Bank

Danya Tazi MokhaCIH Bank