This research proposes a novel approach to language learning through a teaching-based paradigm where learners nurture and instruct AI language companions. We develop a system incorporating three key innovations: (1) Role-Reversal Pedagogy that positions learners as teachers, triggering the protégé effect for enhanced cognitive processing; (2) CEFR (Common European Framework of Reference for Languages)-Based Scoring that dynamically evaluates and adjusts the AI character’s proficiency level; and (3) an Ebbinghaus-Inspired Forgetting Mechanism that simulates realistic memory decay to encourage spaced repetition and long-term motivation. Our implementation demonstrates that modern LLM technologies can effectively simulate language development cycles, with experimental results confirming successful dynamic prompt updating and realistic forgetting patterns that facilitate authentic relearning experiences. This approach addresses critical challenges in language education by enhancing motivation, promoting active engagement, and potentially reducing educational disparities by supporting diverse learning styles. Unlike conventional systems where learners passively receive feedback, our method creates a bidirectional interaction flow that leverages the psychological benefits of teaching to reinforce knowledge acquisition. Future work will focus on educational outcome evaluation, immersive XR implementation, and expansion to broader aspects of language competence beyond grammatical and lexical elements to include pragmatic skills and cultural knowledge.