In this Kaggle competition, we are tasked to build a model that when fed an question answer pair can generate scores that measure subjectively the quality of the pair. We employed pretrained Bert large and Bert base models. Our final submission was an ensemble of these models. Final rank - top 13%.