My research addresses three critical aspects of AI-powered education: pedagogical alignment of LLMs, advanced learner modeling, and intelligent assessment, which directly correspond to three key factors that determine educational learning outcomes: instructional approach, learner understanding, and assessment quality respectively.

In the area of pedagogical alignment, I developed the CLASS framework and code soliloquies technique to ground LLM tutoring systems in established learning science principles. My work in learner modeling includes the question-centric Deep Knowledge Tracing (qDKT) approach and novel frameworks like DUPE and MalAlgoQA, exploring how LLMs can simulate complex student reasoning processes. For intelligent assessment, I created the 'Marking' and Automated Long Answer Grading (ALAG) systems, which provide nuanced, automated feedback on student responses.