Detection of Chagas disease from ECG is largely understudied in computational modelling literature. Moreover, there is a need to detect Chagas from ECG data, as this method is widely accessible and affordable for target populations. Our team, HeartGoesOut2U, proposes a hybrid transformer-CNN model, composed of a pre-trained HuBERT-ECG backbone and Multireceptive Field CNN head. We trained such a model on a merged training dataset, consisting of records from PTB-XL, Code-15%, and SaMi-Trop databases (all downsampled to 100 Hz). The following training pipeline was employed: selected ECG signals were cleaned by removing baseline wander using a Butterworth high-pass filter, utility frequency using a Notch filter, and additional noise using a 4-level DWT. To address the imbalanced nature of Chagas prevalence, positive class weighted Focal Loss was utilised. On the hidden validation set of George B. Moody PhysioNet Challenge 2025, the proposed method achieved a Challenge score of 0.372. Additionally, we conducted an internal evaluation using 5-fold cross validation on public training data. We achieved Challenge score of 0.364, accuracy score of 0.774, and F-1 measure of 0.11. Furthermore, AUROC of 0.785, and AUPRC of 0.143 were observed.