Temples serve as centers of culture and spirituality that encourage prayer, introspection, and civic involvement while reuniting people with their religious roots. They also make major contributions to community development and social welfare. In the field of temple finance management, where transparency is of paramount importance, this research presents a solution which integrates aspects of machine learning and the internet of things. The research offers a novel approach for managing donations made through temple donation boxes that improve security and transparency. By implementing a system that combines real-time image capturing and a hybrid multi-step classification process using convolutional neural networks, Indian currency notes and coins which are dropped into the donation box are classified and their denomination determined. This includes the design and implementation of three unique convolutional neural networks, each having alternating layers of 2D convolutional layers and max pooling layers, with slight modification pertaining to the differences in the structure of coins and notes. A binary classifier, followed by two separate multi-class classifiers, makes the solution end-to-end. Furthermore, the amount deposited, the total amount in the donation box after the depositing of the amount along with the timestamp, which enhances the finance management process’s transparency. While the solution was developed keeping in mind the difficulties in keeping track of donations in large temples, the solution is viable and practical for smaller temples too as the setup process is relatively straightforward and inexpensive. The proposed solution also carries great potential for future development, using deep learning methods providing higher classification accuracy and hardware capable of capturing images of the currency with higher clarity and minimal noise.