Automated Health Coaching: A Study On Integrating Wearable Data With Large Language Models

Fitness is important to sustain well-being and ameliorate the risk of chronic health conditions such as diabetes and obesity. Health coaching has played a crucial role in helping people maintain a healthy standard of living. It has also enabled the betterment of people’s mental well-being due to the role they play in their improved physical activity. The advent of Large Language Models has been a blessing for multiple domains. This work studies the impact of employing well-known LLMs to model the responses of a fitness coach in a coach-participant scenario and how integrating data into LLMs from wearable technologies like smart watches, affects prompt generation. The study is done on a privately aggregated dataset. As described in Gupta et al (2020) , a trained health coach and 28 patients were engaged to collect the first round of human-human health coaching contact via text messages, resulting in the collection of around 2800 texts. The results explore the difference in prompts generated by integrating wearable data into LLMs compared to prompts solely based on conversation data. This thesis investigates the challenges of augmenting these LLMs with wearable data to provide personalized health coaching messages.