This paper describes a novel hierarchical architecture along with its implementation for Credit Unions on credit decision systems with increased interpretability, and performance based on Support Vector Machines (SVMs). The system developed using real credit application data from a Credit Union based in Cuenca, Ecuador partitions the decision process into modular submodels, each focusing on essential financial parameters, risk ratings, and guarantor requirements. Subsequent to modular integration, outputs are passed into an SVM classifier which improves predictive accuracy and robustness. The proposed model achieved greater accuracy than the benchmark traditional flat SVM, Random Forest, and Neural Network as well as attained 83.54% accuracy. Furthermore, the architecture allows financial advisors to audit the system by outputting structured intermediaries in JSON format which enhances regulatory compliance. Because of the modular nature of the model, individual components can be easily updated due to changes in policy or risk parameters adapting to frameworks. This illustrates where adjustable and interpretable AI systems can be implemented into inclusive financial models to assist in reliable credit decision making.