Predicting organic reaction outcomes is challenging because the chemical search space is vast, with even slight context changes capable of significantly altering re-activity. In this paper, we present an AI-powered molecular prediction service de-signed as a deployable component within digital laboratory workflows. Unlike standalone algorithmic approaches, our system integrates a Seq2Seq Transformer architecture with a Beam Search optimizer to predict organic reaction outcomes and function as a decision-support module. The model is trained and evaluated on the USPTO-50k dataset, incorporating standardized preprocessing and a reproducible training pipeline. Experimental results demonstrate strong performance, achieving 97% training efficiency, 57% Top-1 accuracy and improving to 74% Top-10 accuracy with a final loss of 0.1021. We discuss model design, data set preparation, evaluation protocol, and the implications of this work for future AI-driven synthesis planning and automated laboratory systems.