[D] Why does developing these RAG applications feel like alchemy?

\^ Basically the title. Is there a principled way of doing this? Like Weights & Biases, where you can at least monitor what's happening.

applications biases least machinelearning monitor rag rag applications you