Systems and methods are presented for providing a response to a user query. Reception of a user query is detected. An augmentation machine learning model is utilized to determine one or more variations of the user query that correspond to a semantic meaning of the user query. A plurality of response candidates is determined that correspond to the user query by comparing the user query and the one or more variations of the user query to a plurality of documents. A final response candidate is determined from the plurality of response candidates based on utilizing a semantic machine learning model to perform a semantic comparison between the plurality of response candidates and at least the user query.