How we are building the personal health coach

Historically, health and fitness journeys have been fragmented, generic and inaccessible, whether within existing apps or through general health and fitness journeys outside of apps. For instance, a primary care provider might suggest seeing a specialist or losing weight for better diabetes management, but often without providing connections to a nutritionist or fitness coach. This leaves users with the burden of connecting these dots themselves.

Our vision is to address this by offering a proactive, personalized and adaptive AI-powered personal health coach. This coach will seamlessly enable:

• Proactive insights on sleep, fitness and health
• Personalized guidance grounded in behavioral science, established health and wellness principles and individual health metrics, such as activity and other physiological data
• Personalized coaching to set goals and build sustainable habits through actionable, adaptive plans

Starting tomorrow and over the next week, we are rolling out an optional public preview of the health coach for eligible US-based Fitbit Premium Android users and expanding to iOS users soon. Users who opt in to participate will be presented the consent to provide access to their Fitbit data to receive personalized insights.

This innovation is powered by advances in Gemini models plus our new AI-first personal health coach experience on the Fitbit app, and our continuous progress in cutting-edge research across Fitbit, Google Research and Google DeepMind. Building from the ground up takes time and rigor, a commitment especially important for health and wellness. We are approaching this thoughtfully and iteratively, grounded in science, incorporating feedback from the scientific community and users. We will continue to be open about our work through publications and updates, and here we provide a quick peek into what went into building the personal health coach.

“Do I get better sleep after exercising?” sounds like a simple question, but answering it like a proactive, personalized and adaptive coach required several technical innovations.

First, we need the coach to understand and do numerical reasoning on physiological time series data such as sleep and activity, using capabilities similar to those showcased by PH-LLM. For questions like this, the coach verifies recent data availability, chooses the right metrics, contrasts relevant days, contextualizes results against personal baselines and population-level statistics, incorporates prior interactions with the coach, and finally uses the analysis to provide tailored answers and insights.

Second, we utilize a multi-agent framework that coordinates expert sub-agents to provide clear, consistent and holistic support, such as (1) a conversational agent for multi-turn conversations, intent understanding, agent orchestration, context gathering and response generation; (2) a data science agent that iteratively uses tools to fetch, analyze, and summarize relevant data (e.g., sleep and workout data), leveraging code-generation capabilities as needed; and (3) a domain expert, such as a fitness expert that analyzes user data to generate personalized fitness plans and adapt them as progress and context change.

Third, while foundational models are incredibly capable, careful steer is required for it to be useful in the health and wellness context. We developed evaluations based on consumer health and wellness needs to inform the system instructions and improve upon Gemini’s core capabilities in assisting our users.

Working together, these innovations result in more personalized and beneficial guidance.

We know that technical excellence alone isn’t enough, and reliability and safety are paramount.

First, we grounded the coach in scientific and well established coaching and fitness frameworks.

We also used human-centered design to integrate expert and user feedback including:

• Convening a Consumer Health Advisory Panel of leading experts to provide feedback and guidance for the development of the coach.
• Extending the sport science foundation with input with professional fitness coaches to incorporate context-specific approaches.
• Developing novel methods to collaborate with experts, fostering consensus in nuanced areas of personal health and wellness.
• Actively soliciting feedback from thousands of users in large-scale consented research studies (e.g., from Fitbit Insights Explorer, Sleep and Symptom Checker Labs).

Finally, we continuously validate the personal health coach using dimensions of safety, helpfulness, accuracy, relevance, and personalization, collectively known as the SHARP evaluation framework. This multi-level assessment involves over 1 million human annotations and more than 100k hours of human evaluation by generalists and experts across various fields such as sports, sleep, family medicine, cardiology, endocrinology, exercise and behavioral science. This process is further extended and scaled with autoraters to ensure that wellness recommendations are scientifically accurate. The framework also incorporates the coach’s real-world performance, enabling ongoing improvements in the most critical areas.

Join the public preview and share your feedback in the app or through our community forum. You will help shape the coach, so it can do more for and with you.