This paper introduces a novel AI-powered demand prediction framework designed to transform the planning and operation of Demand-Responsive Transit (DRT) systems. Using advanced Graph Convolutional Networks (GCNs) implemented via PyTorch Geometric, the model captures complex spatial-temporal correlations in transit demand across 251 zones in Kalamazoo County, Michigan. By integrating geo-temporal clustering with a Deep Multi-View Spatial-Temporal Network (DMVST-Net), the framework dynamically forecasts passenger requests with high accuracy and geographic specificity. Experimental results demonstrate strong predictive performance, achieving a Mean Absolute Error (MAE) of 4.96 and a Mean Squared Error (MSE) of 32.80 on a dataset of 1.7 million points. More than a forecasting tool, the proposed framework serves as a decision-support system for civil and transportation engineers, enabling data-driven design of DRT service zones, stop placement, fleet deployment, and infrastructure investments. This study marks a significant advancement in bridging deep learning with practical transit planning, offering a scalable, adaptable, and engineering-relevant solution for innovative mobility systems in mid-sized urban regions.