Soorya Ram Shimgekar

I’m a PhD student in Computer Science at the University of Illinois Urbana-Champaign, advised by Koustuv Saha (OnCARE Lab) and Hari Sundaram (Crowd Dynamics Lab).

My research sits at the intersection of natural language processing, computational social science and mental health. I study how people talk about distress online and what happens when large language models become part of those conversations.

My research approaches this from two angles. On the human side, I analyze risk in context using posting histories, digital surroundings, and psychological theories to identify and study distress. On the AI side, I study conversational models to see how they handle vulnerable users, measuring when they drift, refuse help, or amplify dangerous thinking, and pinpointing the internal mechanisms driving those behaviors.

I combine large-scale computational methods with AI safety techniques. My methodological toolkit includes longitudinal social data processing, signed graph neural networks for mapping belief structures, and theory-driven NLP grounded in clinical psychology. To evaluate and understand conversational AI behaviour, I use simulated human behavior, controlled perturbations, and causal analysis, leveraging mechanistic interpretability to uncover models’ internal failure modes. Pairing these techniques with human-centered user studies, I focus on building AI that spots risk earlier, operates safely, and empowers human connections.