GNL Grace
A diabetes educational advisor

Grace runs against a locked evidence base built and curated over more than a decade. Every paper she draws from sits in our Mendeley reference library, exported as a BibTeX file, used as the source of truth. She does not cite from model memory.

Her clinical policies, the cohort framing, the age-band routing, the correction-dose framing, the AID Optimiser positioning, are written down, version-controlled, and published in our open wiki for anyone to read. The retrieval layer pulls from 880 pages of curated content across 7,349 typed chunks, routed by document metadata so the right policy reaches the right question. Phillip Hayes, our Technical Director, built that layer in May 2026.

She is not validated in the regulatory sense of a medical device, because she is not a medical device. She is validated as an educational advisor, against the evidence base she rests on.

I am a paediatric Diabetes Dietitian at Birmingham Women’s and Children’s NHS Foundation Trust, supporting over 300 children and young people with type 1 diabetes and their families. I am also a person living with type 1 diabetes, a husband, and a parent. Grace’s framing, what she covers and how she covers it, is shaped by the questions families ask in clinic, on weekends, and at three in the morning. It is also shaped by the questions I have asked myself.

That is one form of patient and public involvement. We have not yet run a formal service-user advisory panel, and we are open about that. The early review group on the patient and coach side includes Vanessa Haydock and Beth Kelly. On the clinical side, Carmel Smart, Peter Adolfsson, Othmar Moser and others are reviewing against their specialisms. Public feedback channels open on 1 June 2026.

Grace has been live to a closed group since late April 2026. As of early June 2026, over 855 people are registered. The Birmingham Children’s Hospital paediatric diabetes operations team have first refusal ahead of public launch, so the team who taught me how to listen to families get to use her before anyone else does.

Named clinicians across the UK, Europe, Australia and the US hold full-access seats and are testing her against their specialisms, from AID and exercise physiology to paediatric dietetics, alcohol and type 1 diabetes, and psychosocial outcomes. The full picture, who is reviewing what and when, is documented openly in our project repository.

Via Negativa. Clarity by subtraction, not coverage by addition. Grace is built by removing what does not serve the central argument, not by adding more reasons until the argument feels safer.

She is an educational advisor. She is not a clinical decision-support system. She gives population-average estimates at a person’s total daily dose. She does not give a personalised correction dose. She refers every clinical decision to the person’s care team. She uses banned-list language to avoid prescriptive framing.

The theoretical frame for what she is for sits in the work of the diabetes psychology and structured-education communities: that the conversations between people with type 1 diabetes and their clinicians become softer ones about them as a person, not just the numbers. Grace is one tool to help make that happen. She is not the whole of the answer.

Grace only deals with Grades A to D evidence. Grade E content, lived experience, podcast material, FAQ answers, exists in our wiki but is never cited as proof. The wiki backbone is research-grade and feeds her retrieval. The site delivery layer, the voice the reader sees, is evidence-anchored in spirit, not citation-stacked in form. One fully-anchored citation per section is plenty; she carries the depth on demand if you ask for it.

The cohort she rests on is approximately 500,000 patient-days from approximately 1,300 people living with diabetes, run through 77 safety tests across the evidence base. That is the figure used everywhere we describe her, from manuscript to marketing, with no inflation between contexts.

Grace gets a person 80% of the way there with 20% of the effort. The remaining 20% takes self-discovery, guided by human expertise and trial-and-error learning. That is the strapline, and it is the mechanism we make claims about.

We are not claiming a percentage-point HbA1c change. We are not claiming a percentage-point time-in-range change. The evidence for outcomes like that takes years and dedicated trials, and we are open about not having that yet.

What we are claiming, and where we want the evidence to land over the next year, is that conversations about diabetes become softer ones about the person, not the numbers. That families feel less overwhelmed between appointments. That clinicians have more of their limited time available for the human dimension of care. Those are aspirational claims until the data lands. We say so directly.