A neurologic ailment that causes the brain to shrink and brain cells to die, Alzheimer's disease (AD) is progressive. Although there is no known cure for AD, medicines may momentarily lessen or delay the progression of symptoms. As a result, stopping the progression of AD depends heavily on early stage identification. The major goal is to provide a complete framework for early AD diagnosis and medical image categorization for different stages of AD. We used transfer learning to pre-trained models like VGG 16 and ResNet 50 as well as bespoke CNN when using the deep learning technique. There are four categorization metrics in use: Moderate AD might be considered demented, very mildly demented, moderately demented, or not demented. We have developed a web application for remotely assessing and testing AD in order to make it more comfortable for patients and physicians. Based on the AD spectrum, it also establishes the patient's AD stage. In order to classify AD and its prodromal phases, this project incorporates the MRI data obtained from Kaggle. This experiment demonstrates that ResNet 50 and VGG16 have been calibrated to an accuracy level of 95% and 84%, respectively. Also, we created a unique scratch model that classified the 2D multi-class AD stage with an accuracy of 93%.