Effective diabetes management requires ongoing collaboration among patients, providers, and support networks. This paper presents an iterative design and evaluation of a chatbot embedded within a group messaging platform to support team-based diabetes care. Rooted in human-centered AI, the chatbot employs human-in-the-loop methodologies, ensuring continuous input from clinicians, health educators, and patients throughout the design and real-world deployment phases. The system facilitates crucial self-management tasks, including monitoring diet, physical activity, and medication adherence, while offering tailored health education and evidence-based recommendations. By integrating advanced multimodal large language models with a retrieval-augmented generation framework, the chatbot delivers context-aware support that adapts through ongoing human oversight and feedback. Findings reveal its capacity to enhance care coordination. Clinicians valued real-time data insights that inform decision-making, and patients appreciated its seamless integration into everyday communication practices. This participatory approach underscores the importance of involving diverse stakeholders at every stage, boosting usability and trust. Through personalized interventions, the chatbot addresses the complexities of chronic disease management and advances inclusive, team-based digital health solutions. The resulting scalable model significantly demonstrates how human-centered AI can effectively improve outcomes for individuals with diabetes. Ultimately, this research highlights the transformative potential of collaborative health tools across diverse real-world contexts.