This study focuses on integrating machine learning with optical systems, proposing a transformative approach to optimizing laser performance in real-time. By leveraging deep learning, the study aims to generate essential laser beam profiles, such as Top Hat and Laguerre Gaussian, along with their corresponding holograms from scratch. An Autoencoder Encoder Convolutional Neural Network (CNN) model was developed to train the dataset. This model has demonstrated the ability to generate any beam profile, significantly outperforming traditional computational algorithms that require intensive mathematical computations. Traditional techniques take between 10 and 503 seconds to generate a specific beam profile, while the deep learning-based approach can generate any beam profile in just 7 seconds, offering improved performance in terms of time and variety of beam profiles, making it ideal for real-time applications.