<p>Surface electromyography (sEMG)-based recognition of hand gestures is widely used to control robotic arms and prosthetics and assist in rehabilitation. Effective operation of these devices requires intelligent control techniques to ensure precise control. These techniques often use deep learning models, which are profound but extremely sensitive to raw data. Due to the significant randomness in the sEMG signal, designing these models optimally for real-time, subject-specific applications remains challenging. To address this issue, this research provides a framework which integrates a one-dimensional Convolutional Neural Network (1D-CNN) with a Neural Architecture Search (NAS) mechanism. The NAS mechanism employed in the 1D-CNN is enhanced using the LSHADE optimization technique to identify sEMG-based hand gestures effectively, resulting in a framework termed 1D-CNAS. This framework provides an optimized architecture that gives efficient outputs for subject-specific needs. This achieved the highest accuracy compared to the baseline 1D-CNN on data acquired from ten subjects (three female and seven males). Each subject performed six different hand gestures, and in identifying these gestures, the 1D-CNAS achieved a mean accuracy of 80.29%, which is enhanced by 9.57% approx. compared to the base model 1D-CNN. In terms of other mean performance metrics, the precision, recall, and F1-score were 81.13%, 80.29%, and 79.77% respectively.</p>