In a small trial at the Auckland Bioengineering Institute, researchers are testing whether a voice-based AI coach named DiNa can help people with type 2 diabetes keep the promises they have already made to themselves and their clinicians. The difficulty this addresses is not ignorance but the quiet erosion of consistency, since a skipped dose or missed walk seldom brings immediate consequence. Set against a growing population living with the condition in Aotearoa, the work asks a modest question with large stakes: whether steady, personal encouragement between appointments can turn knowledge i
Auckland Researchers Trial DiNa, a Voice-Based AI Coach to Help People With Type 2 Diabetes Stick to Their Plans
DiNa is not about curing diabetes.
Mark
Why build a coach for people who already know what to do?
Mimi
Because knowing isn't the gap. Diabetes can progress quietly, so skipping a dose or a walk costs nothing that day, and habits drift.
Mark
What makes DiNa different from a normal health app?
Mimi
Conversation. You talk to it by voice, and you can photograph meals instead of writing diaries.
Mark
I'd have expected a lifelike avatar to be the selling point.
Mimi
So did the team. Early prototypes had talking heads, but many users preferred voice only. It felt less demanding.
Luke
Let me slow us down. The trial has 10 participants over three months. That can't tell us much about whether DiNa improves adherence or health outcomes.
Mimi
Fair. It's the first of three planned rounds, with larger studies next year.
Luke
And the positive feedback, that monitoring was easier, is what participants told the researchers. That's early impressions, not measured results.
Mark
What about the digital twin part?
Mimi
Early versions model glucose response from carbohydrates, fats and proteins. Personalized meal comparisons are described as something that could eventually happen.
Luke
Right, so that's a goal, not a feature people have today. And whether DiNa works in real health systems is still unknown.
El Pulso
- A condition that advances quietly makes every missed dose and skipped walk easy to excuse, and the gap between knowing and doing keeps widening.
- Self-tracking has become a burden that people abandon, so the team is trying to replace tedious food diaries with photographs, voice corrections and data pulled from devices already in use.
- A three-month trial with just ten participants, split evenly against an existing tool, is the first step in a sequence of rounds that will test whether the idea holds.
- Larger studies are planned for next year, and the parallel digital twin work aims to move the advice toward individual glucose responses, though the results remain preliminary.
In a small trial at the Auckland Bioengineering Institute, researchers are testing whether a voice-based AI coach named DiNa can help people with type 2 diabetes keep the promises they have already made to themselves and their clinicians. The difficulty this addresses is not ignorance but the quiet erosion of consistency, since a skipped dose or missed walk seldom brings immediate consequence. Set against a growing population living with the condition in Aotearoa, the work asks a modest question with large stakes: whether steady, personal encouragement between appointments can turn knowledge into lasting habit, and whether the answer will hold beyond ten participants.
At the Auckland Bioengineering Institute, part of the University of Auckland, a small clinical trial is testing whether an AI coach you can simply talk to can help people with type 2 diabetes do what they already know they should. The coach is called DiNa, and the problem it targets is not a lack of knowledge but consistency. Diabetes management is a long commitment to eating well, moving regularly and taking medication, and because the condition can progress quietly, good intentions often slip under the pressure of work, family and ordinary life.
Dr. Gonzalo Maso Talou leads the team, which sits within the institute's digital twin program. He is clear about the tool's limits: DiNa is not meant to cure diabetes, only to improve the chances that people stick with the plan they built with their clinicians. The need is considerable, with more than 365,000 people in Aotearoa living with diabetes in 2025, and the condition weighs most heavily on Māori, Pacific and Indian communities.
The study follows ten people over three months, half using DiNa and half using an existing tool, so outcomes can be compared. DiNa draws on meal records, medication logs, phone activity and glucose monitors, sparing users from logging every detail. A photo of a meal is enough for the system to estimate its nutritional content, and a spoken correction fills any gaps. The conversation itself is the distinctive feature. Early prototypes showed on-screen faces, but many participants preferred voice alone, which felt more natural and less demanding. The coach can help find a good time to exercise, uncover what is blocking a goal, adjust plans when life intervenes, and point users back to clinicians when medical advice is needed.
The longer ambition is personalised digital twins, computer models of how an individual's body responds to food, exercise and medication. Early versions already model glucose responses to meals, and the team hopes one day to let people compare options or find recipes suited to their current readings. This trial is the first of three planned rounds, and if the results are encouraging, the team hopes DiNa could become widely available within health care.
At the Auckland Bioengineering Institute, part of Waipapa Taumata Rau, University of Auckland, a small clinical trial is testing whether an AI coach you talk to can help people with type 2 diabetes do what they already know they should. The coach is called DiNa. The problem it targets is not a lack of knowledge. It is consistency. Managing type 2 diabetes means a long-term commitment to healthier eating, regular exercise and medication, yet the condition can progress quietly. A missed dose or a skipped walk rarely brings immediate consequences, so good intentions can slip under the pressure of work, family and daily life.
Dr. Gonzalo Maso Talou leads the DiNa team, which sits within the institute's broader digital twin program. The tool is meant to support people between medical appointments and keep them on track with goals set alongside their health care team. Maso Talou is explicit about its limits: DiNa is not about curing diabetes, he says, but about increasing the chances that people stick with the treatment plan they developed with their clinician.
The need is considerable. Health New Zealand's Virtual Diabetes Register counted more than 365,000 people living with diabetes in Aotearoa in 2025, an increase of more than 40% since 2011. The condition disproportionately affects Māori, Pacific and Indian communities.
The current study is small. It follows 10 participants with type 2 diabetes over three months. Half use DiNa and half use an existing diabetes-management tool, so researchers can compare outcomes. The team, which includes Ph.D. student Kekayan Nanthakumar, is asking whether personalized, ongoing support helps people maintain healthy habits more effectively.
DiNa pulls from several sources: meal records, medication tracking, smartphone activity data and continuous glucose monitors. The aim is to spare users from logging every detail. They can photograph a meal, and the AI identifies the foods and estimates calories, carbohydrates, protein and other nutritional information. If something is missing, they can correct or add it by voice. Exercise data comes from phones and wearables already in use, and users can quickly record whether they have taken their medications. Traditional food diaries are time-consuming, Maso Talou says, and people often forget details or stop altogether.
The most distinctive feature is conversation. Users chat or speak with the coach through a voice-only interface. Early prototypes used talking heads on screen, on the assumption that a realistic virtual person would improve the experience, but many participants preferred voice alone. It was less demanding and felt more natural, closer to a phone call than a video call. DiNa can help someone find the best time of day to exercise, identify what is blocking a goal, or adapt a plan when life interferes, offering encouragement when things go well and problem-solving when they do not.
It is not meant to replace clinicians. Its information comes from curated New Zealand diabetes resources developed with clinical partners, and it sends users back to health professionals when clinical advice is needed. With months sometimes separating appointments, Maso Talou says, DiNa can help people understand their condition and the advice they have already received. Early participants told the team the system made monitoring much easier and let them get on with their lives.
The longer ambition is personalized digital twins: computer models simulating how an individual's body responds to food, exercise and medication. Early versions can already model glucose responses from carbohydrate, fat and protein intake. Eventually a user might compare two meals to see which is likelier to spike glucose, or ask for recipes based on current glucose levels and ingredients at home. Maso Talou, who co-leads the twin project with Dr. Finbar Argus, notes that two people can respond very differently to the same meal.
This three-month study is the first of three planned trial rounds, with larger studies scheduled for next year. Those data will also refine the digital twin technology. If results are positive, the team hopes DiNa could be integrated into health care systems and made widely available.
Citas Notables
DiNa is not about curing diabetes; the aim is to increase people's chances of adhering to the treatment plan they've developed with their clinician.— Dr. Gonzalo Maso Talou, Auckland Bioengineering Institute
Quieres la nota completa?
Lee el original en Medical Xpress ↗