The rapid advancement of generative AI and natural language processing has created transformative opportunities for language education, yet the integration of these technologies into pedagogically coherent, CEFR-aligned learning environments remains a significant challenge. This panel presents AI4LANG, a multinational research and development initiative that expands a production-ready AI Language Tutor platform across three typologically distinct languages – English, Japanese, and Croatian – addressing the global need for adaptive, AI-mediated language instruction that serves both high-resource and underrepresented languages.
The platform’s architecture is built on a multi-agent orchestration paradigm where specialised AI agents handle distinct pedagogical functions – exercise generation, pronunciation evaluation, grammar feedback, and learning path adaptation – coordinated through a central orchestrator operating over a dual-database knowledge layer combining vector embeddings with structured lexical and grammatical graphs.
We present the technical architecture, the CEFR-aligned pedagogical framework underpinning adaptive exercise generation, and the methodology for creating language-specific knowledge resources through LLM-assisted inference validated by expert review.