In technology-enhanced learning environments, the integration of Artificial Intelligence (AI) offers new opportunities to address persistent challenges in motivating learners and sustaining engagement. This study introduces “Qinglian Poetry Academy,” an interactive poetry learning application featuring an AI-driven digital human as an embodied pedagogical agent (EPA), designed according to Self-Determination Theory (SDT). To examine the EPA’s effectiveness in meeting students’ fundamental needs and enhancing their learning engagement, a randomized controlled trial was conducted with 420 undergraduate participants. Results indicate that the AI-driven EPA significantly outperformed its non-embodied counterpart in providing perceived support for students’ competence and relatedness. This improvement was accompanied by increased cognitive, behavioral, and agentic engagement, underscoring the EPA’s potential to create a more dynamic and human-centered learning experience. These findings highlight the importance of aligning instructional design with SDT principles when developing AI-driven learning interfaces, thereby contributing to more engaging and responsive educational environments.