How do students in professional fields such as law and forensic science gain realistic practice in high-stakes professional encounters such as client interviews, expert testimony, and cross-examination, when such opportunities are scarce in traditional classrooms? This project develops generative AI role-play simulations that let students rehearse these competencies at scale,…

When students have an AI tutor at their fingertips, what do they actually ask, and does it actually help them learn? This project examines this question at scale, using fine-grained interaction data from NUS’s introductory programming courses to understand how AI tutors are really used, and how their use can…

When students integrate generative AI into their academic lives, what emerges over time? LLM Trajectories follows undergraduates across multiple NUS faculties to trace how patterns of LLM use take shape across a degree, combining periodic surveys with detailed analysis of student–AI conversations to examine not just what students say about…

This 2024–2025 research addresses LLM-as-a-judge “agreeableness bias” (TNR < 25%). We propose a minority-veto strategy and regression-based framework using human-annotated data to ensure reliable, scalable, and unbiased evaluation of LLM-generated content.

This 2024–2025 research at IIT Kanpur explores a hybrid model where Codaveri AI augments Teaching Assistants. This human-in-the-loop approach provides scalable, scaffolded programming feedback, ensuring high-quality instruction and enhanced lab efficiency.