Tea leaves, which come from Camellia sinensis, are essential to world economies, but their productivity and quality can be impacted by diseases like grey blight (Pestalotiopsis spp.) and blister blight (Exobasidium vexans). The subjective and error-prone nature of traditional eye inspection for illness identification is what motivates the use of machine learning. Using a Kaggle dataset of 885 photos, this study classifies eight tea leaf disorders (such as red leaf spot and healthy) using convolutional neural networks (CNNs)—a proprietary 5-layer CNN, ResNet50, and MobileNetV3. Rotations and flips are examples of data augmentation that improve CNN performance by growing the dataset and lowering overfitting. Confusion matrices confirmed that the custom CNN had an accuracy of 91.40%, ResNet50 92.74%, and MobileNetV3 93.55%, with a few minor misclassifications resulting from visual similarities. Applications include precision farming, resource optimization, outbreak prediction, and mobile diagnostics for farmers. Generalizability is limited by issues such as data scarcity. To enhance automated tea disease control and promote sustainable production and global economic stability, future initiatives should give priority to a variety of datasets, generative learning, and multimodal techniques.