An artificial intelligence-powered platform called INSULAIT, designed to identify insulin resistance and prediabetes early and provide personalized lifestyle guidance, successfully automated the interpretation of metabolic laboratory data and delivered tailored recommendations during preliminary testing, according to a research abstract presented at the 2026 AACE Annual Meeting in Las Vegas. 

According to the abstract, which was led by Mihail Zilbermint, MD, MBA, FACE, of Johns Hopkins Medicine in Baltimore, researchers built the system on an open-source LLaMA large language model and aligned it with reinforcement learning using human feedback informed by expert-reviewed literature, clinical guidelines, and curated educational materials focused on insulin resistance and metabolic health. 

INSULAIT integrates fasting glucose, insulin levels, lipid profile parameters, hemoglobin A1c values, and continuous glucose monitoring data into a proprietary algorithm that generates an INSULAIT score representing prediabetes risk. Patients either upload laboratory reports as PDF files or images or manually enter laboratory values into structured fields. An artificial intelligence (AI) coach then delivers real-time, evidence-based lifestyle suggestions. 

According to the researchers, during preliminary testing, the platform accurately extracted and synthesized metabolic laboratory results from uploaded documents and correctly mapped manually entered values into the risk algorithm across diverse user profiles. Dynamic changes in INSULAIT scores corresponded with trends in glucose markers, behavioral inputs, and patient-logged lifestyle data. 

High Engagement, No Inappropriate Advice

User testing showed high engagement, with most participants completing the risk assessment, successfully uploading or entering laboratory values, and interacting with at least one feedback module. Users also reported improved understanding of insulin resistance and rated the AI-generated lifestyle recommendations as clear, actionable, and relevant, particularly for nutrition, physical activity, sleep hygiene, stress management, and behavioral barriers. Expert review identified no inappropriate advice or safety concerns during the pilot phase. 

The researchers noted that the study was limited to preliminary feasibility and usability testing and did not include larger clinically characterized cohorts or comparisons with physician-led assessment. They added that future studies will evaluate diagnostic performance, metabolic outcomes, behavioral changes, and progression to prediabetes or type 2 diabetes. 

Disclosures: Several members of the research team were affiliated with INSULAIT and POWERUP DIGITAL. Funding information was not reported.