SourceDetect utilizes Keras and Tensorflow to train and apply a convolutional neural network model to perform a transient event search through TESS data. The code scans imaging cutouts, computes detection likelihoods, and classifies candidate point-like events as positive, negative, or artifacts and outputs tables of detections including positions, classes, and event confidences. SourceDetect is optimized for large-scale surveys, handling vectorized image inputs and providing programmable filtering criteria to isolate transient, variable, or moving sources.