
We discuss the problem you have in detail with an aim to get to a solution. With My experience in increasing language model accuracies, writing async LLM as a judge scripts, designing complex RAG engines and LLM model frameworks, redesigning memory management in codebases and pre/post training, I am pretty sure I can give you a solid headway if not resolve the problem entirely in these 30 mins.
I might wave the fees off if you're an interested student who has done their homework.