APPLYING RAG IN EVALUATING, GIVING FEEDBACK AND HELPING USERS IMPROVE THEIR IELTS TEST PERFORMANCE.

Abstract

In recent years, English certificates (e.g., IELTS) have become increasingly important as enterprises, both national and international, now view them as essential requirements rather than mere advantages when recruiting new employees. This has led to an explosion in the number of IELTS learners and a significantly increased demand for improving test performance, especially in the Writing skill. Writing is widely considered not the hardest skill, but the most challenging one to improve, as both the weaknesses and their corresponding solutions vary significantly among examinees. Unfortunately, IELTS Writing training classes often come with high fees for personalized and accurate guidance from professionals, while free resources and online materials tend to address only parts of the problem. Therefore, in this thesis, we introduce a complete IELTS Writing assessment system that provides band scores, identifies strengths and weaknesses, and generates an improvement roadmap based on the given topic and essay. The system supports both IELTS Writing Task 1 and Task 2, ensuring that all outputs are closely aligned with the actual content of the user’s writing. To achieve this, we leverage a large language model – Gemini 2.5 Flash – within a RAG (Retrieval-Augmented Generation) architecture, utilizing a diverse and comprehensive collection of IELTS Writing-related documents. Furthermore, all technologies and techniques employed in this system, such as the Gemini 2.5 Flash LLM and the Weaviate vector database, were carefully researched, benchmarked, and compared against alternative options to ensure that the final choices best meet the project’s requirements. Overall, this study contributes to the development of intelligent educational tools by demonstrating the effectiveness of a RAG-enhanced approach for automated IELTS Writing evaluation. The system highlights how integrating retrieval with generation can offer a scalable and reliable solution for personalized writing support.