<p><?tk 4?>Accurate prediction of pharmacokinetic parameters is critical in drug discovery, yet traditional experimental approaches are time-intensive and costly. This study presents a real-time artificial intelligence framework that integrates graph attention networks, Transformer models, and automated machine learning to predict pharmacokinetic parameters, including those related to absorption, distribution, metabolism, and excretion. Unlike static models, our dynamic approach periodically incorporates newly available data, stratified by administration routes such as intravenous and oral, and recalibrates model parameters without full retraining. This flexibility enables the system to integrate new compounds from single-point measurements while maintaining high predictive accuracy over time. The optimized models achieved a mean coefficient of determination of 0.93 and a mean absolute error of 0.059, demonstrating improved performance compared to conventional batch learning techniques. Despite challenges such as computational costs and model interpretability, the proposed framework shows significant promise in enhancing scalability, responsiveness, and prediction accuracy, highlighting its potential to accelerate data-driven decision-making in drug development.</p> Graphical abstract <p></p>