Smart wristbands enable users to control technology through subtle hand gestures by decoding muscle signals. However, each individual's muscle signals are unique, making personalization a critical challenge. Leveraging a publicly available dataset (Kaifosh et al. 2025), our team developed a personalized gesture recognition model using surface electromyography (sEMG) data from 100 participants performing 9 gestures. Addressing inter-user variability through within-user training, we engineered 160 features and selected 37 via random forest ranking and correlation pruning. Logistic regression with L2 regularization achieved strong cross-validation performance (F1 Macro = 0.7164), but holdout testing revealed a generalization gap (F1 Macro = 0.3977). Performance varied widely, confirming heterogeneity in performance across diverse users. Future work could explore adaptive time windows and fine-tuning pre-trained models to enable more robust commercial neuromotor interfaces.