Chagas disease, endemic in several regions of Latin America, is characterized, in its chronic phase, by electrocardiographic alterations that are often subtle and difficult to detect clinically. In the context of the George B. Moody PhysioNet Challenge 2025, this study presents a machine learning-based approach for binary classification of Trypanosoma cruzi infection from digital 12-lead ECG recordings. Data from the SaMi-Trop collection were used, with traces composed of 4096 samples per derivation. The signals were extracted from files in HDF5 format, vectorized by concatenating the derivations and organized into stratified training and testing sets. Modeling was performed with AutoGluon TabularPredictor, which allows automated evaluation of multiple architectures. The highest performing models – RandomForest (Gini criterion) and a weighted ensemble – achieved an accuracy of 57.8%. The confusion matrix and qualitative analysis indicated limitations resulting from interindividual variability of signals and the absence of morphological normalization. Nevertheless, the results demonstrate the feasibility of AutoML approaches as preliminary tools for computational screening of Chagas cardiomyopathy, especially in resource-limited settings. Future work may incorporate morphological feature extraction strategies, model interpretation methods (e.g. SHAP) and architectures based on deep neural networks to improve accuracy and generalization.