Prototype for building structured RAG: could this work? Hi everyone, I’ll start by saying that I have a humanities background and a passion for programming, but only recently have I started getting closer to AI and its underlying structures. During my studies, I noticed that certain structures could be assimilated to linguistic-psychological models and translated into algorithms. I started some extra study sessions brainstorming with AI: the "notes" in the GitHub repo are the result (please note that the form and exposition are AI-generated; I only needed the content and source references to dive deeper). From there, it was a short step to creating a prototype using vibecoding. # The Project The idea focuses on the targeted creation of RAG based on the tokens of user-written prompts, in order to provide the language model with targeted documentation and, possibly, without noise. To provide the necessary knowledge, we use graphs based on language structure (AST).…