DESIGN AND EVALUATION OF A RETRIEVAL-AUGMENTED AI TUTOR FOR ACADEMIC ENGLISH WRITING IN SAUDI HIGHER EDUCATION

Resumen

There has been growing interest in the role that academic English writing plays in assessment, publishing, and employment opportunities for initiatives that aim at improving national capability in Saudi Arabia [29,30]. However, in parallel with growing popularity of use of generative AI for writing amongst students, validating this AI raises challenge as it can produce unlimited writing fluently with either false references or overwrite the authoring process while masking their logical thinking behind elegant writing. Retrieval-Augmented Generation (RAG) is a technical remedy as it entails generation of text by retrieving knowledge from the source corpus which can be more readily auditable [3,4]. This research paper provides a review of existing literature published in the years 2020-2025 covering various aspects of RAG architecture, retrieval evaluation, LLM adaptation, and academic integrity governance towards designing and evaluating an AI tutor based on the RAG technology for academic English writing. In compliance with PRISMA guidelines [1] for the design-synthesis methodology of systems [2], a methodology strategy adopted for this purpose encompasses failure mode analysis vis-à-vis system controls and performance indicator analysis resulting in (i) the reference architecture consisting of authorized corpus, hybrid retrieval and re-ranking, schema awareness in tutoring, quality gates, and monitoring and (ii) two synthesis tables of design criteria and evaluation standards along with deployment requirements.