Self-monitoring of blood glucose is essential for individuals with diabetes mellitus, yet conventional methods remain invasive and often cause pain, discomfort, and risk of infection. These limitations highlight the need for alternative approaches that maintain accuracy while improving user comfort and reducing waste. This study develops a non-invasive blood glucose monitoring system that integrates computer vision, laser-based imaging, deep learning, and mobile health technology. The prototype consists of a Raspberry Pi, a red laser module, and a high-resolution camera designed to capture standardized images of the fingertip. A total of 152 individuals participated in data collection, and after quality evaluation, 83 valid datasets were retained for analysis. Image preprocessing and dataset refinement were performed to ensure consistency, followed by model training using the MobileNetV2 architecture. Oversampling and controlled data augmentation were applied to address class imbalance and improve model generalization. The system’s prediction output was connected to the Android application “Laras NIDDS” through Internet of Things integration to enable real-time monitoring. Performance evaluation produced an accuracy of 85 percent, a recall of 90 percent for high blood glucose detection, and an area under the curve value of 0.92, reflecting strong discriminative capability. These results indicate that laser-assisted imaging combined with deep learning provides a feasible non-invasive method for estimating glucose levels. The system offers practical advantages, including improved comfort, reduced procedural risks, and elimination of disposable testing materials. Although clinical validation with larger and more diverse samples is required, the findings demonstrate the potential of this approach to support continuous and sustainable self-monitoring. Furthermore, the developed system contributes to the advancement of digital health technologies and presents a promising alternative for routine glucose assessment. Further analysis showed that consistent finger placement and controlled illumination significantly improved image quality, supporting the reliability of the captured optical patterns. The integration of hardware and software components also demonstrated stable operational performance, indicating that the system is suitable for continuous home-based use. Future development will focus on enhancing model precision, expanding dataset diversity, and refining the mobile application interface to improve usability. Overall, the system represents a practical step toward accessible non-invasive glucose monitoring.