Chagas disease remains severely underdiagnosed due to expensive serological testing that is largely unavailable in endemic regions. We present an automated screening sys- tem that transforms standard 12-lead ECG into an accessi- ble diagnostic tool using ResNet-based deep learning with demographic fusion. Our approach addresses three criti- cal challenges: expensive diagnostic infrastructure, lack of specialized expertise in endemic areas, and extreme data scarcity for model training. The system employs residual blocks for multi-scale tem- poral ECG analysis, capturing disease-specific conduc- tion delays and morphological changes. To ensure clinical safety, we implement uncertainty quantification through Monte Carlo Dropout, providing conservative predictions when model confidence is low. We address severe class imbalance through weighted loss functions and controlled synthetic data augmentation, enabling robust training de- spite limited positive samples. Our solution achieves physionet evaluation of 0.245 while maintaining real-time inference (<100ms) on stan- dard CPUs, representing a 10-200× cost reduction com- pared to traditional diagnostics. This work demonstrates the first deep learning system specifically engineered for Chagas detection in resource-limited settings, combining clinical safety through uncertainty estimation with practi- cal deployment considerations for endemic regions where the disease burden is highest.