Recently, STACK (System for Teaching and Assessment using a Computer Algebra Kernel) has significantly advanced mathematics education by enabling symbolic feedback to students. However, the high complexity of authoring its contents becomes a barrier for teachers to start to use, where it requires knowledge of Computer Algebra System (CAS) programming. Generative AI can offer a potential solution for this limitation, although it may ”hallucinate” incorrect generations, due to the stochastic nature. In this study, we present a design of an Automated Question Generation (AQG) tool using generative AI. It adopts Retrieval-Augmented Generation (RAG) to ensure curricular alignment and an external CAS-Maxima engine to validate the generated code deterministically. Besides, it utilizes Depth of Knowledge (DOK) framework to manage cognitive complexity. By validating the symbolic logic before serialization, this solution ensures that the resulting XML files for STACK templates are error-free and ready for deployment.