Many experts in AI predict that AI will transform education, particularly through personalized learning. But what does this look like in real educational settings, especially within the structure of a classroom? How can AI-driven personalization be integrated meaningfully into language instruction without undermining the role of teachers or peer collaboration?

This paper introduces a new AI-enhanced instructional framework called Fluid Language Pedagogy (FLaP), designed to complement (not replace) classroom-based language learning. FLaP supports a hybrid learning environment in which students engage with AI chatbots and AI-generated materials tailored to their individual proficiency levels and goals, while still participating in classroom interaction, group work, and teacher-guided activities. The “fluid” aspect of FLaP refers to its adaptable structure: learners can move between self-directed AI-supported practice and collaborative classroom engagement, allowing for flexible learning pathways.

A core feature of FLaP is the use of customized large language models (LLMs) for different learner levels. We developed novice and intermediate LLMs for Japanese learners using prompt engineering—a process easily adaptable to other languages. This paper details the structure of FLaP, the development of level-specific LLMs through prompt writing, and its classroom implementation through a sample syllabus and integrated learning activities.