This work aims to develop an algorithm for the automatic detection of Chagas disease using 12-lead ECG. To that end, we adapted a residual network, initially trained in a South American population to estimate the patient biological age using the ECG. To perform this adjustment, we used Low-Rank Adaptation to update the weights of the convolutional layers of the networks. To mitigate labeling uncertainties, we trained several models using different sub-datasets , which are part of the challenge public training set. These models were then assembled to provide a prediction on the presence of Chagas disease from the ECG. Locally, the dataset was split into a 90/10\,\% training/internal test sets stratified by demographics, chagas labels and origin of the database to optimize the hyperparameters. Hyperparameters that provided the best internal validation challenge score were used to train the final model in the full training set. Our model achieved a challenge metric score of 0.263 on the internal test set and of 0.259 on the external validation dataset (team ChagAI). The results highlight the difficulties of detecting Chagas disease, even with a large training dataset. Despite a better than random performance, on internal testing and external validation sets, the model requires further optimization to improve its performance if it is intended to be used in clinical practice.