Research Projects
These projects describe the broader research programs behind my papers: the questions we ask, the systems we build, and the educational problems we aim to solve.
How can a tutoring agent recognize that a student's unexpected step is still correct?
This project evaluates LLM feedback in propositional logic using knowledge-graph-derived ground truth. It distinguishes optimal steps, valid alternatives, and incorrect reasoning so that feedback quality can be measured where adaptive tutoring matters most.
Does adding a second agent actually make tutoring feedback more reliable?
This work studies role-specialized tutoring pipelines in which one agent generates feedback and another verifies it. The project investigates how upstream quality, problem complexity, and agent roles affect the final pedagogical response.
What do teachers and students need for LLM tools to support—not displace—learning?
This research synthesizes evidence about student-facing LLM tools and develops teacher-informed guidance for classroom adoption. It connects technical design decisions with pedagogy, assessment, and responsible implementation.