DEVELOPING AN AI-POWERED ENGLISH LEARNING PLATFORM WITH ADAPTIVE LEARNING PATH DESIGN

Abstract

This thesis addresses the challenge of limited personalization and adaptability in current English learning platforms, particularly within listening and speaking practice. Traditional systems depend on static materials and lack automated evaluation, restricting their capacity to deliver individualized learning experiences. To overcome these limitations, this study proposes an AI-powered platform integrating Groq for primary speech-to-text transcription, Whisper as fallback processing, GPT-4o-mini and Azure Speech for text-to-speech generation, and Azure pronunciation assessment. DeepSeek is incorporated to automatically generate structured speaking content, including dialogue scripts, vocabulary, and grammar points based on selected topics and proficiency levels. A BPMN-based dynamic learning-flow module enables flexible instructional sequencing for both administrators and learners. The system supports dictation, role-play, and shadowing activities, offering real-time feedback on content accuracy and pronunciation quality. Experimental results demonstrate improved adaptability, reduced lesson-creation workload, and consistent evaluation performance. The study concludes that multi service AI integration combined with workflow-driven design can enhance personalized and scalable English-language learning.