Keyword spotting (KWS) is the task of detecting specific keywords in audio streams and plays a critical role in various applications in the speech domain. This paper presents a multilingual keyword-spotting system for educational applications, particularly speech and hearing-impaired classrooms. We have targeted four languages, namely Hindi, Bengali, Odia, and English, and thirteen classroom instruction-related keywords for each language. We created a dataset containing 1564 training samples employing eight speakers for development. We build the system using an RNN-BiLSTM architecture and use Mel-frequency cepstral coefficients (MFCCs) to extract features. A test dataset of 180 samples is used to test the system, and we achieve a classification accuracy of 97.35% with an average confidence score of 98.65%. This system could bridge communication gaps in inclusive educational environments.