Mangoes, known as the “King of Fruits,” hold a central role in the agricultural and economic landscape, especially in regions like Pakistan, where varieties such as Anwar Ratool, Chaunsa, and Langra are widely cherished. Their diverse taste profiles—ranging from sweet to sour and tangy—are key factors influencing consumer preferences. However, traditional methods of identifying mango varieties and predicting taste rely heavily on subjective human assessment, often leading to inconsistencies, errors, and increased post-harvest losses. In this project, we propose TasteNet, a novel deep learning-based solution utilizing Convolutional Neural Networks (CNN) for the dual task of mango variety classification and taste prediction through image analysis. The primary goal is to develop a robust and accurate model that aids in improving agricultural practices, minimizing food waste, and helping consumers choose mangoes based on taste preferences. The CNN architecture comprises convolutional layers, pooling layers, optimized using TensorFlow. Data augmentation techniques were applied to enhance model generalization and address challenges like varying image angles and lighting conditions. Evaluation was conducted using confusion matrices and training curves. The CNN architecture was designed with dual output layers to simultaneously predict mango variety and taste. The model achieved impressive accuracy in both tasks during validation, successfully distinguishing between mango types and taste categories in complex real-world scenarios. Future development will explore broader applications, integrating predictions for other fruit types and enabling real-time deployment for commercial and consumer-facing use cases. The research emphasizes the transformative role of artificial intelligence in modernizing agriculture and enhancing food selection processes.