The AccelNet-MultiNet Fellowship has come to a close for several program participants, and among them is Asia Maurich Novelli, a first-year PhD student at the BRAN Lab at NetSI London, supervised by Prof. Andreia Sofia Teixeira. This past spring, Asia traveled from London to Boston to spend a month working with Prof. Christoph Riedl and postdoctoral researcher Sean Kelley on a project integrating AI, mental health, and network science.

The AccelNet-MultiNet Program, supported by the National Science Foundation, creates structured opportunities for graduate students, postdoctoral researchers, and early-career faculty to cross institutional and national boundaries and conduct research on multilayer networks at partner institutions on both sides of the Atlantic. The program is built on the premise that some of the most important scientific advances happen not within a single lab, but in the productive friction between different research cultures, methods, and perspectives.

For Asia, the motivation to participate was both scientific and strategic. Her PhD sits at the intersection of mental health, network science, and machine learning, with a growing focus on human–AI interaction and its implications for wellbeing. As AI tools for companionship and emotional support become part of everyday life for millions of people, she saw an urgent need to understand how these interactions shape behavioral patterns and mental health trajectories, but to pursue that question rigorously, she needed new skills. Asia came to Boston hoping to learn how to construct synthetic data through LLM-LLM simulation, and to turn her still-nascent research idea into a concrete experimental framework.

The project she developed with Prof. Riedl, Dr. Kelley, and Prof. Teixeira asked a pointed question: does a chatbot respond differently depending on the mental health profile of the person it is talking to, and does that behavior shift across conditions, severity levels, and over the course of a conversation? To investigate this, the team built an LLM-LLM simulation framework in which language-model-generated personas — each defined by clinical profiles — held multi-turn conversations with a chatbot operating without a system prompt, simulating a real-world scenario. Working with synthetic personas only at this stage, it also allowed the team to vary severity across conditions including anxiety, depression, eating disorders, and psychosis, while avoiding the ethical risks of exposing vulnerable individuals to unmonitored AI interactions. Asia led the end-to-end design of the framework, from profile generation to the analysis pipeline, examining how the chatbot's responses differed across profiles in both their semantic content and in patterns of refusal versus engagement.

Asia returned to London with a working experimental framework, preliminary results, and a sharper sense of where the project goes next. The collaboration with Prof. Riedl and Dr. Kelley is ongoing, with new research objectives already taking shape. Beyond the immediate outputs, the experience reinforced something Asia believes more broadly: that being physically embedded in a new research environment — encountering different ways of structuring problems, communicating ideas, and navigating uncertainty — makes for a more adaptable and more intentional researcher.

‍