Kerala’s Deaf and Hard-of-Hearing population primarily communicates through Malayalam Sign Language (MSL). However, the spatial-temporal complexity of MSL motions is beyond the scope of current recognition methods. Utilizing its ability to extract spatial and temporal characteristics from video input, this paper explores the use of 3D Convolutional Neural Networks for MSL recognition. Unlike traditional 2D CNNs, which process frames independently, 3D CNNs analyze entire motion sequences leading to improved gesture recognition accuracy. The model is trained on a custom MSL video dataset, where each sign is represented as a continuous motion sequence. To improve model performance, several preprocessing methods are used such as frame extraction and normalization. Superior recognition accuracy results from the 3D CNN model’s efficient learning of spatial-temporal relationships. Recall, accuracy, precision, and F1-score are used to assess performance, showing notable gains over traditional 2D CNN techniques. This research contributes to AI-driven assistive technology for MSL recognition, aligning with the Sustainable Development Goals (SDGs) and the Indian Knowledge System by promoting inclusive and digital accessibility for the Deaf community.