When generating candidate summaries against job descriptions, how does your AI safeguard against potential bias crept into the training of the model? Or how do you address AI hallucination ?