Compared to offline learners, online learners' attitude during the learning process is relatively poor, and a feeling of loneliness is entailed as they often study alone. This results in a low learning outcome. So far, no examples exist for the design of a learning companion to this end. Herein we present a pioneering work on a co-existing, artificial learning companion capable of improving the learner's attitude through sleepiness detection. We capture, analyze and estimate the level of sleepiness employing a machine learning technique with the pilot study data. Then, we propose a prototype called LearniCube using a sleepiness detection model with an experimental evaluation of LearniCube.