The direct binary prediction with LLMs for coding seems impossible as a single erroneous bot would break the application
I guess rethinking the entire stack, system level, hardware could solve this problem.
when you create a custom NN using nn.Module, you are basically using COMPOSITION on the .__init__() of that NN.
COMPOSITION: an object that OWNS the object(s) of another class. owns-a relationship as they say.