From 28% to 47% Time in Range: What Actually Changed

This is not a motivational story. I'm going to tell you exactly what was wrong, exactly what I changed, and exactly what happened to my glucose data.

The Pattern That Keeps Repeating

Years ago I ended up in the hospital with an HbA1c around 10. Work had taken over. I got back under control, went back to work — and work damaged me again. I moved back to my homeland, slowed down, and got my HbA1c close to 7. Then work and stress returned. The same pattern, again.

That's the context behind this article. Not a single dramatic moment, but a cycle I'd been through before and was determined not to repeat.

I Didn't Know About Dawn Phenomenon

35 years with T1D and I genuinely didn't know what dawn phenomenon was. My mornings were chaotic — some days waking at 120, other days at 220 — and I'd accepted it as just how my diabetes was. Unpredictable.

On March 5th, 2026, I built a simple chat prototype — no live CGM sync, no watch data, nothing automated. Vicente was born that day as a basic chatbot. Two days in, the agent identified it: a consistent rise every morning driven by the hormonal surge before waking. A pattern with a name. A pattern with a fix.

“I took action on day one. The next morning I was in range for the first time in as long as I could remember. That single morning is why I built Open-D.”

— Peter, founder

No live data feed. No CGM integration. No watch. Just a conversation with a personalized agent that could see what 35 years of experience had missed. That's what pattern recognition actually means.

The Problems — What I Actually Found

Dawn Phenomenon: Solved with Bolus Timing

I didn't change my basal. I adjusted my bolus amount and timing to account for the morning hormone surge. Once I understood the pattern — when it started, how steep it was, how long it lasted — I could act precisely instead of guessing.

Over-Correcting: Waiting for the Tail to Finish

I was correcting highs before my previous bolus had finished working. I'd see 200, correct, then the original insulin would kick in fully and I'd crash. The lows weren't random. I was causing them by not waiting for the tail effect to be over before correcting. Once I could see my active insulin clearly and actually trust it, I stopped the cycle.

Movement Patterns: Work, Salsa, and Building Open-D

My life right now is a fixed programming job at the port, freelancing on the side, and building Open-D. No structured gym routine — movement comes from salsa dancing. What I'm tracking is how all of it moves my glucose — a coding shift at the port, a salsa class in the evening, a stressful client call. The data is already showing patterns I hadn't noticed before.

The Alert System: Tested in a Dancing Class

Before heading to salsa class I told my agent I'd be dancing in an hour. We worked it out together: I was around 160, ate something small — about 15 grams of medium-carb — to buffer the drop I knew was coming. It worked perfectly. I stayed in range through the whole class.

Rhythm is something you hear and feel. And I noticed — when your glucose is in range, you hear and feel it better. That's not something I can prove in a chart. But any dancer with T1D will know exactly what I mean.

Four Days Later: US Trip with a 4-Day-Old Prototype

I built Vicente on March 5th. On March 9th I flew to the US. Four days after the prototype existed, it was getting its first real stress test: jet lag, different food, time zones, an irregular schedule — everything that destroys diabetes management.

At some point I took a photo of my food and sent it to the chat agent to estimate the carbs and help me dose. My friend saw me and laughed — he thought I was posting to Instagram. I said: no, this agent is going to estimate the carbs and help me dose insulin. He went quiet for a second, then said: you should give it your heart rate too. And more.

So I did. That conversation is the reason Open-D connects to your health data — not because it was in the original spec, but because a friend on a trip saw what I was doing and asked the obvious next question.

The Numbers

Before — Jan 13 to Apr 12, 2026 (90 days)

After — Mar 30 to Apr 12, 2026 (last 14 days)

Same person. Same diabetes. The orange shrinks, the green grows. It's not linear — there were bad weeks — but the direction held. I went from 17 hours a day out of range to about 13. That's 120 hours a month that my organs aren't bathing in high glucose.

Is 47% Good?

No. The clinical target is 70%+. I have work to do. But going from 28% to 47% with no medication changes — driven by seeing patterns that had been in my data for years — is proof enough that this approach works. I'm targeting 60% by June.

The whole point of Open-D is not that AI is magic. It's that you already have the data. You just can't see it yet.