AI + Levels: Understand your continuous glucose monitoring data with a simple AI stack.

Levels collects continuous glucose data via a CGM, typically worn for 14 days at a time, paired with your food and activity logs. The app visualises your glucose responses to specific meals, exercise, and sleep, calculating a 'Metabolic Score' and providing zone recommendations. Within the app, you see detailed glucose curves, time in range, glucose variability, and estimated HbA1c. You can log food manually or by importing from other apps, and attach photos. Activity and sleep data integrate from wearables. For export, Levels allows users to download their raw glucose data, food logs, and activity records. Typically, these are available as CSV files, containing timestamps, glucose values, and associated metadata. While the in-app experience provides immediate feedback and structured insights, the exported data is a rich source for deeper, personalised analysis beyond the app's predefined metrics.

Our approach involves a four-part stack to transform raw Levels data into refined insights. At the foundation is Levels itself, serving as your primary data source, capturing the granular details of your metabolic responses. Layered on top is your chat assistant, such as ChatGPT, Claude, or Gemini. This tool becomes your primary point of interaction for immediate questions and pattern recognition within your weekly data exports. You can ask it to summarise trends, identify anomalies, or explain specific glucose responses. Next, your notebook tool, like NotebookLM, acts as a long-term memory for your health journey. Here, you store all your Levels data exports, chat assistant summaries, and your own reflections, allowing for extended context and synthesis over months. This forms your personalised 'Ledger' in our Research → Ledger → Protocol methodology. Finally, an agent layer, which could be a simple scheduled action or a custom GPT, orchestrates periodic tasks. This layer ensures you receive prompts for data export, consistent review reminders, and compiled summaries, moving you from raw data to actionable 'Protocol' steps without manual overhead.

Establishing a consistent weekly ritual is key. Designate a specific day, perhaps Sunday afternoon, as your 'data review day'. First, export your glucose, food, and activity data from Levels for the preceding week. Consolidate these into a well-organised folder. Next, open your chat assistant and paste in your current week's data, using a prompt designed to identify patterns, highlight significant responses, and compare against your goals. Review the assistant's output, noting any surprising findings or persistent trends. Transfer these summary insights, along with the raw data, into your notebook tool. This builds a cumulative record of your metabolic responses over time. Journal your reflections directly in the notebook tool – how you felt, what adjustments you made, and what you plan to experiment with in the coming week. If you identify persistent concerns or patterns that require expert interpretation, compile a concise summary from your notebook tool, ready to share with your healthcare practitioner or coach.

It is crucial to understand the limitations of this AI stack. This method will not provide medical diagnoses, nor should it ever replace the professional opinion and guidance of qualified healthcare practitioners. AI tools are pattern recognition engines; they do not possess clinical judgment or an understanding of your full medical history. This stack is not designed for closed-loop dosing or self-prescription of medication based on glucose readings. It will not automatically adjust your insulin or make definitive claims about specific foods or interventions without your critical oversight. Its purpose is to help you observe, reflect, and structure information for more informed discussions with your healthcare team, not to practice medicine itself. Always consult with a doctor for any health concerns or before making significant changes to your diet, exercise, or medication regimen.