This paper presents an innovative approach to enhancing foreign language proficiency through the integration of large language models (LLMs) within adaptive instructional systems (AIS). Focused on the Department of Defense (DoD) language training mission, we detail the development of a proficiency-driven pipeline for Modern Standard Arabic. Our approach combines personalized scheduling of task-critical vocabulary (TCV) with scaffolded linguistic exercises to support students’ progression from basic recall to higher-order language skills such as reading comprehension and translation. By leveraging LLMs for real-time content generation and translation evaluation, we deliver targeted, actionable feedback to students while providing instructors with data-driven insights into individual learning trajectories. The structured feed-back and scoring metrics ensure continuous improvement in language competency while addressing challenges in automated grading and user trust. We conclude with lessons learned and future directions for refining our LLM-based AIS pipeline.