Comment by ghita_

That's cool! How does it perform compared to more "naive" methods? How did you go about comparing that performance, and was it in a real world RAG?