Multitask learning (MTL) is an advanced approach in machine learning that tackles multiple tasks concurrently, leveraging shared representations to enhance overall performance. This method proves especially beneficial in Natural Language Processing (NLP), where tasks often exhibit common structures and semantics. In this study, we investigate the implementation of multitask learning using three variants of the BERT model: BERT uncased, ALBERT uncased, and RoBERTa uncased. Our focus is on two distinct tasks: Query Correctness and Sentiment Analysis. We provide a comparative analysis of these models, highlighting the superior performance of RoBERTa uncased when these tasks are performed jointly, thereby establishing it as the most effective model for our specific multitask learning setup.