A Generative-AI Platform for Developing Health Coaching Agents with Simulated Users

Health coaching (HC) supports sustainable lifestyle behavior change through personalized, collaborative interactions but is labor-intensive and time-consuming, making it difficult to scale. Generative artificial intelligence (AI) could extend this capacity alongside human health coaches. Building such systems, however, requires testing and evaluation with a diverse group of clients who differ in goals, readiness, and communication style, which is costly and potentially risky to carry out with real users. Here, we present an inspectable platform that pairs a modular health coach agent (HCA) with a configurable simulated user agent (SUA) to enable iterative testing of AI-driven health coaches in silico. The platform allows researchers to define documented client cases, vary coaching competencies and model backends, run automated HCA–SUA conversations, and inspect the strategy, state transitions, and safety decisions behind each turn. It also supports HC training by providing human trainees with access to practice sessions with multiple AI-simulated users. Our demonstration illustrates configurable interaction and turn-level observability rather than clinical effectiveness, providing a controlled setting for developing health coaching agents before human evaluation.