Zakat is a fundamental pillar of Islamic finance that serves as a mechanism for wealth redistribution. However, there is currently no Indonesian-language Question Answering System (QAS) capable of automatically and contextually responding to zakat-related queries. This study aims to develop a zakat-focused QAS using a Long Short-Term Memory (LSTM) model integrated into the Telegram platform. The dataset was compiled from the official BAZNAS zakat guidebook and processed through tokenization, padding, and label encoding. The model architecture consists of an embedding layer, two stacked LSTM layers (with return sequences, dropout, and recurrent dropout), followed by two dense layers (200 and 100 units) with additional dropout layers before the softmax output. The model was trained using the Adam optimizer (learning rate 0.003), a batch size of 24, and 100 epochs. Evaluation was conducted using a confusion matrix, resulting in a validation accuracy of 93%, with a precision of 0.94, recall of 0.93, and F1-score of 0.92 (weighted average). The system was deployed via the Telegram Bot API and demonstrated response times under two seconds, with stable performance across hundreds of question labels. This work contributes to the advancement of digital zakat education and presents a scalable solution that can be further extended within the ecosystem of Islamic Finance Technology and Digital Religious Education.