Specially abled children often face difficulties in expressing their emotions and receiving proper responses from their environment. Convolutional neural networks (CNNs), a deep learning approach, is integrated in a table-top robot model in conjunction with MediaPipe and OpenCV for live and spontaneous real-time gesture detection to improve accessible communication. The multimodal interface facilitates smooth communication between caregivers and children by filling in communication gaps. It improves the level of communication for children who face difficulty in expressing themselves verbally by capturing hand movements and translating them into visual and audio responses. Caregivers have expressed satisfaction with the early testing, which yielded encouraging results with high accuracy in gesture recognition. This system is a flexible tool that promotes accessibility and interactivity, as it can be tailored to different contexts.