Student engagement is crucial for successful self-paced learning. Feeling isolated during self-paced learning with neither adequate supervision nor intervention by teachers may cause negative emotions such as anxiety. Such emotions may in turn significantly weaken students' motivation to engage in learning activities. In this paper, we develop a self-pacedlearning environment (FishBuddy) that aims to reduce anxiety and promote student engagement. We construct and implement a physiologically-state-aware performance-evaluation model for identifying potentially fruitful moments of intervention when students show frustration during learning activities using an Apple Watch application that measures heart rate and alerts the student to watch a visualization of his or her own physiological state. We have conducted an experiment with 20 first-year undergraduate students, randomly separated into an experimental group and a control group, who carry out online, self-paced English grammar exercises. The students in the experimental group used FishBuddy and those in the control group did not. The self-reports from both groups show that FishBuddy significantly reduced reported experiences of anxiety and isolation in the experiment. Further to this, students who used FishBuddy were engaged longer with the exercises. The average scores on the exercises between the two groups, however, were not significantly different.