Your Profile Is Not Your State

Before health AI recommends what you should do next, it should understand what appears to be true about you now.

Most health personalization starts with a profile: age, conditions, medication context, goals and preferences. That information matters because it describes who you are, but it does not necessarily describe the state you are in today.

Sleep changes overnight. Activity changes during the day. Appetite can shift by lunch, and symptoms can appear by the afternoon. Your profile may look almost identical on Monday and Thursday even when your body does not.

Two sides of the same patient

I experienced this gap personally while using GLP-1s and managing metabolic health and prediabetes. The care plan could remain relatively stable while the daily decisions I had to make did not. Some days felt normal; others brought different appetite, tolerance, energy or routines.

That experience shaped an idea we now spend a lot of time thinking about at DrunR: profile and state are two sides of the same patient.

The profile is the slower-moving side: diagnosed conditions, medication context, goals, preferences and history. State is the faster-moving side: recent meals, activity, sleep, appetite, symptoms and other signals that may help explain what is happening now.

When available, wearables can add changes in heart rate or HRV compared with the person's own baseline. A consented CGM can add glucose patterns. DrunR's current state architecture is designed to bring signals like these together rather than asking any one datapoint to explain the person.    Call Transcript Summary

A system that sees only the profile can therefore know a great deal about someone and still return essentially the same recommendation every day. That is personalization on paper, not necessarily personalization in practice.

The goal is not to monitor everything. It is to create a useful, time-specific mirror from the evidence that is available, understand what is missing, and then decide how much confidence to place in the recommendation.

Take recovery as one example. In our Core Intelligence Layer, recovery need reflects recent activity, sleep, changes in heart signals from the person's own baseline, and how they were recovering before. No single wearable number decides the answer; several signals contribute context together.

The same idea applies elsewhere. Food tolerance, for example, can depend much more on today's symptoms and recent experience than on a permanent label. These states are designed to influence recommendations; they are not diagnoses or measures of clinical severity.

What this means in real life

Now picture the same GLP-1 patient looking at the same Grilled Salmon Bowl twice in one week.

On Monday, they slept well, report no current symptoms and have a normal appetite. The full bowl appears to be a strong fit.

On Thursday, almost everything in their profile remains the same: same patient, goals, medication context, preferences and salmon bowl. But sleep was poor, appetite is lower and mild nausea is present.

[Image: Monday and Thursday screens side by side. Caption: “Illustrative DrunR product concept showing how the same meal may fit differently under different patient states.”]

The salmon bowl has not suddenly become “bad,” and the patient has not become a different person. What changed is how well that meal fits that person at that moment.

So the better answer may not be another restaurant or a banned food. It may simply be the same bowl adapted: less rice, sauce on the side, a smaller portion.

This is why DrunR separates the food from the patient's current state before producing a recommendation. The meal remains its own object. State represents what appears to be happening with the person now. Only then does the system ask how well the two fit.

State also has to keep moving. Today's decision, response and new signals can contribute to tomorrow's understanding. At the same time, nobody should need a wearable or CGM for the system to work. Missing information should remain missing, and confidence should decrease rather than being replaced with false certainty.

Understanding state is not diagnosis. It is simply a more disciplined question to ask before AI speaks:

What does the available evidence suggest may matter for this person right now, and how confident are we?

Knowing who the patient is matters. Understanding where they are right now is what can make personalization genuinely personal.

‍