Type 2 diabetes mellitus (T2DM) has a significant impact on health, requiring effective dietary management to control blood glucose levels. Adherence to dietary recommendations is a challenge, especially in developing countries such as Peru. This study presents a deep learning model that aims to provide personalized dietary guidance for patients with T2DM. Using multilayer perceptron (MLP) neural networks, the model accurately estimates meal amounts and ingredient proportions, addressing variability in meal compositions and cultural dietary habits. A dataset of 300 cases with 50 different Peruvian dishes was created with the help of a specialized nutritionist. Two MLP models were developed to predict ingredient proportions and total portion sizes, evaluated using metrics such as MSE, RMSE, MAE and R \(^2\) score. The MLP models outperformed other algorithms, demonstrating the potential to improve dietary management of patients by providing personalized recommendations to improve the care of patients with this disease.