In this paper we propose the development of an online movie recommender system, based on information extracted from user interaction with a Web application that allows creation of personal watchlists. Because we want the system to work in near-real-time mode, we will use prediction models generated and optimized by Automated machine learning (AutoML) frameworks, for which limited time budgets are set. AutoML provides methods and processes to build machine learning (ML) models with high scale, efficiency, and productivity all while sustaining model quality. The construction of ML pipelines will be made with two state-of-the-art AutoML tools: H2O AutoML and Auto-sklearn. The comparative analysis aims at the accuracy of the prediction models but also at fitting into the allocated time budget. To evaluate and compare the results of the prediction models, we additionally used our own model, based on the XGBoost algorithm, optimized without a time limit. The obtained results confirm that each of the two AutoML frameworks can be selected as a solution for integration in the recommender system, due to the capabilities of generating optimized prediction models in a short time.