We have several synthetic datasets and automated evaluation options for such things that were close to impossible to do before LLMs.