Online education is increasingly popular, but one challenge many teachers face is making their lectures engaging and interactive. Structured oral debates can be an effective way to achieve this, encouraging critical thinking and learning among students. However, scalable and reproducible methodologies for automatically generating such debate—particularly in Spanish—remain limited. This paper presents a modular approach for generating podcast-style educational debates in Spanish utilizing Large Language Models (LLMs). The proposed pipeline combines academic text summarization, structured prompting, and multilingual adaptation to produce topic-grounded, pedagogically meaningful conversations. The methodology is validated through a multidisciplinary case study that spans five academic domains and is evaluated using BLEU scores across various LLM configurations, including zero-shot and fine-tuned open-source models. Results show that high-quality, coherent debates can be generated even in low-resource settings, supporting the development of inclusive, AI-powered educational tools for Spanish-speaking learners.