Rural communities in Northern Florida bear a disproportionate burden of obesity, prediabetes, and type 2 diabetes, driven by structural barriers including limited access to specialty care, provider shortages, transportation burdens, and broadband gaps that constrain the reach of even telehealth interventions. At the same time, artificial intelligence AI-enhanced tools offer a scalable pathway to extend evidence-based obesity and diabetes prevention to underserved settings.

Professors Bonnie Spring, Ph.D., Hongyu Miao, Ph.D., and Assistant Professor Xiaonan Zhang, Ph.D., collaborated across FSU colleges to develop a virtual lifestyle management coach designed to help expand access to lifestyle management and overcome barriers to care. They initially trained the virtual coach on thousands of anonymized coaching calls from Spring’s prior weight loss trials.

They then upgraded the agent’s knowledge with training on Motivational Interviewing, Nutrition, and Exercise Science. Most AI-based conversational agents are developed for short, question-answer exchanges. In contrast, a live behavioral coach builds a long-term supportive relationship across many weeks of coaching sessions.

The relationship, called a “therapeutic alliance,” requires keeping track of a person’s goals as they evolve over time, in addition to their unresolved barriers to change and the stable principles that guide behavioral counseling.

The substantial demands on different memory systems cause most virtual agents to lose track of a person’s detailed history over extended use, resulting in mistakes and repetitive questioning. To overcome these challenges, the team’s virtual coach consists of multiple agents with defined roles that work as a team.

A summary agent maintains a persistent coaching state aligned with SMART goal principles. An automated signal generator agent integrates this guiding state with recent chat history to generate a compact steering instruction. These are injected into a dialogue agent’s prompt for response

By separating stable counseling principles from dynamic session guidance and assigning different responsibilities to a team of collaborating AI agents, the system supports future digital health coaching systems that require proactive agenda management based on reliable continuity, goal tracking, and follow-up.