Everybody talks about RAG (Retrieval Augmented Generation) or one of the many other variants. But how do you prepare your documents to make them available to generative AI models? And more specifically, how do you split your documents into chunks? Is semantic chunking really the best way to do that? In this talk, we will see a novel technique to prepare your documents along with the methods to validate its quality.
We will also explore the subtle but extremely important differences of the Embedding models and how to use them effectively to improve the quality of the Semantic Search, which is the basis for most of the RAG techniques.