This research paper presents an in-depth exploration of a neural network architecture tailored for intent classification using sentence embeddings. The model comprises a feedforward neural network with two hidden layers, ReLU activation functions, and softmax activation in the output layer. This paper meticulously examines the technical intricacies involved in data preprocessing, model architecture definition, training methodologies, and evaluation criteria. Detailed explanations are provided for the rationale behind architectural decisions, including the incorporation of dropout layers for regularization and class weight balancing techniques for handling imbalanced datasets. Moreover, the mathematical foundations of the chosen loss function (sparse categorical crossentropy) and optimization algorithm (Adam optimizer) are thoroughly elucidated, shedding light on their roles in facilitating model training and convergence. Through empirical experiments and theoretical analyses, this paper offers insights into the effectiveness and resilience of the proposed neural network architecture for intent classification tasks. It serves as a technical guide for engineers aiming to comprehend, implement, and optimize neural network models for practical application in natural language processing endeavors.