AI + Hedia: how to integrate your CGM companion app into a personal health data system.

Hedia serves as a central hub for diabetes management, primarily focusing on insulin dose calculation, carb tracking, and glucose logging. The app allows users to manually input carbohydrate intake, record insulin doses (both bolus and basal), and log continuous glucose monitoring (CGM) readings, often integrating directly with some CGM devices for automatic data transfer. Beyond active data entry, Hedia also tracks physical activity and general lifestyle notes, providing a comprehensive daily record. While Hedia offers in-app visualisations like glucose trends and insulin-on-board, the true value for deeper analysis lies in its data export capabilities. Users can typically export their logged data, including glucose readings, insulin doses, and carb entries, in formats such as CSV or PDF. These exportable files represent the raw material for your personal AI stack, moving data from a confined app interface into a format where it can be queried and analysed systematically. What remains largely within the app are immediate alerts and real-time guidance; the historical, granular data is what interests us for stacking.

To genuinely make sense of your Hedia data, we advocate for a structured, multi-tool approach. This stack begins with Hedia itself, serving as your primary data collection point. Hedia provides the raw, sequential information about your glucose levels, insulin dosages, and carbohydrate intake. The second layer is your chat assistant (like ChatGPT, Claude, or Gemini). This tool becomes your immediate analyst, capable of processing the exported Hedia data to identify patterns, answer specific questions about weekly trends, or summarise key metrics. Your notebook tool (such as NotebookLM) comprises the third layer. This is where you store all your exported data over time, alongside the summarised reports from your chat assistant and any personal observations. Critically, the notebook tool offers long-context windows, allowing it to synthesise insights across weeks or months of data, far beyond what a single chat assistant interaction can manage. Finally, an agent layer – akin to scheduled actions, custom GPTs, or workflow tools – provides the automation, ensuring you receive regular nudges, summary emails, or protocol reminders based on your accumulated data and insights. This layered approach aligns with our 3-Layer method: Hedia provides the raw Ledger data, your chat assistant and notebook tool facilitate Research, and the agent layer helps implement specific Protocols.

Establishing a consistent weekly ritual is key to extracting sustained value from your Hedia data. Designate a specific day, perhaps Sunday evening, as your 'data review' slot. Your first step is to export the past week's data from Hedia. Ensure the export is in a machine-readable format, such as CSV. Next, upload this CSV file into your chat assistant along with a specific prompt designed to highlight relevant weekly metrics and trends. The chat assistant will provide an initial summary, identifying glucose variability, insulin usage patterns, and carbohydrate distribution. Review this summary carefully. Take any notable insights, questions, or anomalies identified by the chat assistant and add them to your notebook tool, alongside the raw data export itself. This creates a growing, searchable record of your health journey. If the summary flags persistent issues like unexplained high variability or consistent hypoglycaemic events, document these for review with your healthcare practitioner. The goal is not to self-diagnose but to arrive at appointments with structured, longitudinal data and specific observations.

It is crucial to understand the limitations of this AI stack. Firstly, it will not provide medical diagnosis. The insights generated are analytical interpretations of your data, not clinical assessments. Any health-related decisions must always be made in consultation with a qualified healthcare professional. Secondly, this stack does not replace the expertise of your endocrinologist, diabetologist, or diabetes educator; rather, it empowers you with better information for those consultations. Thirdly, it is not a closed-loop automated insulin delivery system, nor does it recommend changes to your insulin dosages or medical treatment plan. The stack processes your historical data; it does not offer real-time medical interventions. Finally, while it can highlight patterns, it cannot interpret complex individual physiological responses or account for unlogged variables in the way a human clinician can. It is a support tool, not a substitute for professional medical care.