My current work centers on LLM-based affective computing and human–AI interaction. I am interested in understanding why a model behaves as it does, especially in emotionally sensitive settings. Improvements in aggregate scores are useful, but I am more interested in the behavioral changes behind those improvements and whether they can be explained in a meaningful way.
I also work on spatio-temporal forecasting. In this area, I focus on incorporating structured information into predictive models while keeping the evaluation protocol clear and reproducible. This work has made me particularly attentive to whether an observed gain reflects a genuine modeling contribution.
More broadly, I am drawn to research questions that begin with a simple observation and become more interesting once tested. I enjoy problems where careful experiments can gradually turn an empirical pattern into a clearer understanding of model behavior.
Some of this work is still ongoing. I will expand this page as the projects become ready to share publicly.