Indian classical dance forms, a profound expression of India’s cultural heritage, employs intricate hand gestures known as “mudras”, encompassing both unilateral (Asamyukta) and bilateral (Samyukta) movements. These gestures function as visual language and possess the capacity to articulate nuanced narratives and symbolic meanings. This study introduces a novel Deep Convolutional Neural Network (CNN) model based on the lightweight MobileNet-V2 architecture for mudra classification in videos. The proposed model incorporates a streamlined CNN architectural blueprint infused with depth-separable convolutions, which has been designed with dual aims: (i) drastically reducing computational overheads and (ii) ensuring swift, effective real-time video classification. Trained on curated dataset comprising open-source videos and images, the proposed model’s achieves a classification accuracy of 98.58% with an average inference time of 68 ms. The results demonstrate the model’s potential as an invaluable tool for enthusiasts seeking automated recognition of mudras in Indian classical dance.