The miniaturization of wearable electrocardiogram (ECG) sensors is revolutionizing the monitoring and detection of arrhythmias by enabling the monitoring of individuals at scale. Although wearable ECG recorders can alleviate the shortage of ECG recorder hardware, there arises a need to aid clinicians in interpreting the vast amounts of data collected by these devices. In this study, we propose an approach to apply a state-of-the-art metric learning algorithm while addressing the practical limitations and challenges encountered when deploying it in local clinical environments in Malaysia. In our experiments, we propose modifications to the sampling and data preprocessing strategy, achieving a sizable performance improvement over the baseline performance.