using litellm - for now shuffle disabled

now using 4.1mini with temperature 1

removed parameters from eval script

modified prompts for ask question and build personas - fixed temperature to 0 and model to 4.1 nano for testing

fix on answers

added function to save also users

added variability in synth answers

modify n of fleet in test script

complete eval flow ready

using direct embedding comparison

added pairwise approach

added eval script and data