• Valeria R. Mendoza

  Author

• Alejandro M. Quispe

  Author

This paper introduces a unified approach to combine artificial-intelligence-based adaptive assessment with large language model (LLM) tools for the assessment and improvement of the English language proficiency of university students. Current writing instruction and proficiency assessment are mostly rigid, with fixed sequences of items and standardized feedback that are not attuned to the variable content of individual students' learning processes, their cognitive burden, or the conventions of writing in their subject matter. This research integrates empirical findings from the fields of AI-based proficiency assessment, automated essay grading, and generative-AI as a measure of writing feedback, leveraging the advancements of the computerized adaptive testing (CAT) approach and recent transformer-based language models, to create a two-level architecture comprising an Adaptive Assessment Engine and an LLM-Based Academic Writing Support Module, both linked by a learner model to promote personalized learning. The Adaptive Assessment Engine automatically adjusts item difficulty and skill coverage in real time based on ability estimation and the writing support module provides rubric-aligned, register-sensitive and iterative feedback on structure and coherence, lexical sophistication, and grammatical accuracy. The design rationale and evaluation criteria for the framework are informed by a narrative synthesis of 17 peer-reviewed studies published from 2021 to 2025. In addition to the documented improvements in scoring reliability, engagement, and revision productivity noted in the literature, there are ongoing questions about validity, algorithmic bias, over-reliance, and issues in academic integrity. The paper concludes that it is possible to integrate an adaptive-assessment and writing-support architecture design that can improve formative and summative assessment of English proficiency while still maintaining the oversight of the instructor, and it provides directions for empirical validation, cross-linguistic generalization, and longitudinal impact studies of university English-language programs.

• PDF