We are excited to introduce our innovative proprietary large language model tailored specifically for women’s health, now entering testing in Oura Labs. This model, integrated into the Oura Advisor experience, leverages a wealth of established medical knowledge, overseen by board-certified clinicians and women’s health experts. It analyzes biometric signals and long-term trends to provide personalized, evidence-based health guidance.

This model covers a comprehensive range of reproductive health topics, from menstrual cycles to menopause, representing a significant shift towards more personalized, clinically-informed AI applications for women's health. Ricky Bloomfield, MD, Oura's chief medical officer, emphasized that traditional one-size-fits-all AI approaches fail to address the complex and often overlooked aspects of women’s health. Oura's model sets a new benchmark by combining trusted clinical science with real-world biometric data, enhancing both the accuracy and relevance of health advice provided to women.

As women increasingly seek AI guidance for issues like cycle irregularities and perimenopause, the need for context-aware models grounded in clinical understanding becomes vital. By referencing curated research and analyzing individual biometric data, the Oura Advisor equips users with critical insights, promoting informed and confident conversations with healthcare providers.

Moreover, the model is designed with privacy in mind; all interactions are securely hosted on Oura's infrastructure, ensuring conversations are not shared or used for external AI training. Participants in Oura Labs will help refine this model, contributing to a collective understanding of women's health that promises to enhance support for cycles, fertility, pregnancy, and hormonal health, underlining Oura's commitment to responsible AI deployment.

K Health is revolutionizing patient care by refining its AI physician, Gemma 3, using real-world clinical data to improve conversational capabilities. By leveraging a 24/7 virtual primary care platform, K Health provides accessible services such as urgent care, chronic condition management, medical weight loss, and mental health support.

To enhance patient intake processes, K Health migrated its AI model to Gemma 3 on Google Cloud's Vertex AI, which reduced inference costs while boosting service personalization. The existing intake chat system proved less conversational and empathetic, prompting the team to aim for an interaction style that felt more professional.

The solution involved employing smaller, well-tuned AI models over larger ones, prioritizing decision-making logic training rather than simple content generation. After thorough evaluations, the team selected Gemma 3 for its balance of performance and cost. They implemented a structured training framework using multi-node GPU clusters, allowing for efficient scaling of model variants.

Direct preference optimization (DPO) was utilized to refine conversation quality based on accuracy, coherence, and clinical outcomes like referrals or prescriptions. The DPO training notably improved the Gemma 3 4B model’s performance score from 0.48 to 0.76, affirming the team's hypothesis that smaller models, when finely tuned, can outperform larger domain-specific variants.

Additionally, the team utilized Axolotl AI with Accelerate in a custom multi-node setup, significantly slashing training time by two-thirds while achieving 90-95% model accuracy. These innovations reduced API call numbers per chat, leading to increased efficiency and lower operational costs. As a result of these advancements, K Health’s AI physician now offers a more natural conversational experience, underscoring the potential of AI in enhancing clinical interactions and patient care outcomes.

Two years after leaving Google, Fitbit co-founders James Park and Eric Friedman have unveiled their new venture: Luffu, an AI-driven service designed to monitor family health. This initiative builds on their past success in personal health tracking, aiming to facilitate shared health management and alleviate the mental burden of caregiving.

Luffu is described as an "intelligent family care system," currently in private testing with around 40 employees, many of whom are former Google and Fitbit staff. The service utilizes AI to consolidate fragmented family health data, recognize routines, and identify significant changes before they escalate into health crises. Users can log health information through voice, text, or photos, with the AI efficiently extracting details and detecting patterns among family members, including pets.

Key features empower users to ask questions such as, "Is Dad's new meal plan impacting his blood pressure?" or "Did someone give the dog his medication?" The AI responds with customized answers and visual representations, providing targeted insights while prioritizing user privacy. Alerts are tailored to reduce anxiety without unnecessary oversight, ensuring families feel supported rather than monitored.

The inspiration behind Luffu stemmed from Park and Friedman's personal experiences in managing care for aging parents, revealing the shortcomings of existing tools designed only for individuals rather than family groups. With approximately 63 million adults in the U.S. serving as family caregivers, the need for such a comprehensive, user-friendly system is increasingly evident.

Looking ahead, while Luffu intends to broaden its scope beyond software into hardware, the immediate focus remains on refining the app and onboarding initial users, paving the way for a more connected and responsive caregiving experience.

Omada Health recently introduced OmadaSpark, a groundbreaking AI nutrition assistant developed in collaboration with AWS and Meta, aimed at enhancing virtual healthcare delivery. Launched in 2025 and built on AWS infrastructure, OmadaSpark utilizes a fine-tuned Llama 3.1 model on Amazon SageMaker to provide real-time motivational interviewing and personalized nutrition education. This innovative solution helps users tackle motivational challenges like emotional eating and supports them in making healthier food choices.

OmadaSpark's core feature, Nutritional Education, dramatically simplifies access to nutritional guidance, reducing the time needed to answer dietary questions from days to mere seconds. The AI agent offers personalized feedback based on individual member data and past interactions, ensuring relevant advice is provided. This real-time support encourages users to discover and implement healthier eating habits while fostering a positive relationship with food through tools like barcode scanning and photo recognition for meal tracking.

The close collaboration between Omada’s clinical and AI teams ensured that the model met real-world healthcare needs while upholding high standards of care. Rigorous testing protocols, including human reviews of output, guarantee the clinical accuracy and safety of the information provided.

Following the implementation, member engagement surged; those using OmadaSpark were three times more likely to return to the app than non-users. By leveraging advanced AI technology, Omada Health not only enhances the role of health coaches but also paves the way for ongoing developments in care delivery, benefiting patients with timely and evidence-based nutritional education. As they look toward future enhancements, Omada remains committed to ethical AI use in healthcare, ensuring a human-centered approach at scale.