A complete system for managing, monitoring and guiding type 1 diabetes.

Glucose Mentor on a laptop and a phone: the doctor page and a low glucose alert in Telegram, with demo data.

Every image on this page shows a fictional patient with demo data.

I built Glucose Mentor for myself first. I live with type 1 diabetes and wanted a system that watches with me around the clock, understands what I eat and gives my doctor the full picture at a glance.

Today it runs every day. It reads the continuous glucose monitor (CGM), talks with me in Telegram and keeps a dashboard that my doctor and I look at together. Doctors who have already seen it have received it very warmly.

Three things, one system

Managing

You write, say or photograph what you eat in Telegram. The system understands plain language, estimates the carbs and keeps the meal and dose diary by itself.

Monitoring

It watches the sensor around the clock. It warns before a low, keeps insisting on an urgent low until you answer and even remembers when the sensor expires.

Guiding

An AI mentor answers in plain words from your own data: why did I go up, what did that meal do, what pattern do my nights follow.

The mentor in Telegram

Day-to-day contact happens in one Telegram chat, like talking to someone who knows you well. A few moments from an ordinary day:

Telegram chat: a predicted low at 81, an urgent alert at 69 with the buttons "I took sugar" and "Sensor error", a juice, and a check 15 minutes later at 92. — Low glucose: first the prediction, then the urgent alert with one-tap buttons, until you are out of the low.

Telegram chat: a photo of lentil soup and bread, an estimate of about 50 grams of carbs, the log entry and a two-hour check at 158. — Meal photo: it estimates the dish and the carbs, logs it with one "yes" and tells you how it went two hours later.

Telegram chat: a steep rise to 186, the question "Did you eat something?", the answer "a sesame bread ring" and the log entry. — An unexplained rise: it asks what you ate and fills in the diary by itself.

Telegram chat: the question "why do I go up every morning?", an answer about the dawn phenomenon based on 14 days, and a note kept for the doctor. — Asking the mentor: it answers from your data and keeps for the doctor whatever you ask it to.

Telegram chat: the sensor expires in 2 hours, a sensor change, quiet during warm-up and advice for the first 24 hours. — The sensor: a reminder before it expires and extra care in the first hours of a new one.

It also sends meal reminders, asks about the dose when a meal has none, corrects times and portions from one sentence and warns if the phone stops sending. The low alarm also runs on the phone itself, even when the computer is off.

The dashboard

All the data ends up in a dashboard with large text, in dark and light themes. Each tab answers one question.

The "My week" tab: seven small glucose charts, one per day, each with its time in range. — Seven days side by side on the same scale, each with its time in range. One tap opens the day with its meals.

The "Today at a glance" tab: the current glucose value and the curve of the last 24 hours with meals. — The current value with its trend arrow and the 24-hour curve with the meals on it.

History

The key numbers, the typical 24 hours of all days (AGP) and the hour-by-hour profile. This is also where you ask your own history.

The "Learning" tab: meal response, nights and the accuracy of the warnings. — What the system has learned about you: how you respond to meals, how your nights go and how accurate its warnings turn out.

The "Doctor" tab: time in ranges, key numbers and the month calendar. — The report for the doctor: the picture at a glance, the course over time and detailed charts that open when needed.

For the doctor: questions in plain language

The doctor opens their own page with a code and asks the way they would ask a colleague: "How have the nights been over the last 14 days?", "What happens after evening meals?".

The answer arrives in 10 to 18 seconds, as cards with the key number up front and up to three charts drawn from the data itself. It answers only from the patient’s data and says clearly what is missing. What would take digging through exports and reports is in front of the doctor in seconds.

The whole page prints as an A4 report. Anything the patient asked in Telegram to keep for the doctor appears under "Selected cases", with the meal, the curve and the conversation.

The question "How have the nights been over the last 14 days?" and the answer as cards with a chart, with demo data. — A question from the doctor and the system’s answer, with the chart of the nights.

The safety rules

A system that talks about blood sugar needs firm limits. These apply to every answer:

Insulin doses are decided by you and your doctor. The system records your doses and shows them to the doctor.
On a low, glucose comes first: the instruction is always fast carbs now.
While a low lasts, only the alarm speaks. Questions about food wait.
Low warnings run around the clock, and their limits change only after testing against the history.
Every answer rests on your data. For studies, the mentor searches and gives the source.
For every treatment decision, the doctor has the final word.

How it is built

The system runs on the patient’s computer and updates a private, code-protected page for the doctor. The mentor uses Anthropic’s latest AI models (Claude) and talks in Telegram.

It is built around one person: their hours, their food, the targets they set with their doctor. That is exactly what Luram does: AI systems shaped around the people who use them.

Are you a diabetologist?

If you would like to see it up close, get in touch: Contact [email protected]. I will show it to you live, with demo data.