<p>The rapid expansion of quick commerce platforms like BlinkIt, Zepto, and JioMart has introduced unique challenges in understanding customer sentiments due to their operational focus on ultra-fast deliveries and hyper-local logistics. This study conducts a comprehensive analysis of sentiment classification methodologies, exploring both traditional ML techniques and advanced DL models to classify customer reviews into positive, negative, and neutral categories. Traditional models, while offering simplicity and interpretability, achieved moderate accuracy (83% with SVM) but struggled to capture the complexities of neutral sentiments. In contrast, DL models, particularly LSTM, achieved superior performance with an accuracy of 88.96% and a macro F1-score of 0.64, leveraging pre-trained embeddings like GloVe to enhance semantic understanding and contextual representation. Further experiments with optimizers, including Adam, RMSprop, SGD, and Nadam, revealed their limited impact on resolving class imbalance and improving neutral sentiment classification. To address these challenges, we integrated hybrid architectures combining GloVe and BERT embeddings, achieving a significant accuracy of 90.69% and demonstrating improved generalization across sentiment classes. However, the classification of neutral sentiments remained a persistent challenge, underscoring the need for advanced techniques like data augmentation and ensemble strategies. This research highlights the importance of adopting hybrid and deep learning-based approaches for sentiment analysis in quick commerce platforms. The findings provide actionable insights for enhancing customer satisfaction and service quality, while also paving the way for future research in domain-specific sentiment classification and scalable solutions for underrepresented sentiment categories.</p>