IELTS Master English AI: An AI-Powered IELTS Preparation Platform with LLM-Based Assessment, Multi-Metric Pronunciation Scoring, and Adaptive Spaced Repetition

Tóm tắt

Automated assessment of subjective English language skills remains an open challenge in computer-assisted language learning. This paper presents a comprehensive preparation platform for the International English Language Testing System (IELTS) built on an event-driven hybrid architecture that separates artificial intelligence inference from core application logic via asynchronous message brokering. The platform implements three principal techniques: large language model based writing and speaking grading with structured rubric output aligned to official IELTS band descriptors, multimodal speaking assessment combining local speech-to-text transcription with direct audio analysis for pronunciation and fluency evaluation, and adaptive spaced repetition vocabulary scheduling with bounded interval growth. Preliminary validation on 30 humanscored IELTS essays demonstrates strong inter-rater agreement between the automated grader and human examiners, achieving a Pearson correlation of 0.97, a quadratic-weighted Cohen’s kappa of 0.96, and 96.7% of predicted scores falling within half a band of the human reference. A dedicated multi-metric pronunciation scoring pipeline achieves strong error-severity discrimination with a Spearman rank correlation of −0.957. Stochastic simulation further verifies that the spaced repetition implementation produces bounded, adaptive review intervals that correctly differentiate across learner proficiency levels.