Haythron update
Building EndoMind's Foundations Before the Product
Why EndoMind began with traceability, organised clinical knowledge, clear information boundaries and independent evaluation before a polished interface.
Starting somewhere less visible
The visible part of an AI product is usually the exciting part. A user enters a question, an answer appears and the project suddenly feels real.
It would have been easy to begin EndoMind with that demonstration. We deliberately started somewhere less visible.
Before building a polished interface, we worked on information identity, traceability, knowledge organisation, evaluation and reproducibility.
As a clinician, I did not only want to see whether EndoMind could produce a fluent answer. I wanted to know where the information behind it had come from, how it had been interpreted and whether the reasoning could be revisited later.
Keeping knowledge in context
Clinical knowledge evolves. Recommendations change, information is revised and conclusions may differ depending on the population, setting or question being considered. Even an accurate statement can lose its meaning when separated from its original context.
EndoMind therefore needed clear traceability and clear boundaries between different kinds of knowledge. My specialist teaching is different from external clinical information. LLM-generated material should remain identifiable as such rather than quietly becoming either of them.
This was not paperwork added around the product. It was part of the product.
A practical foundation
We also began building a clinical knowledge map. Its purpose was practical: to organise disease mechanisms, phenotypes, investigations, interventions and outcomes in a form that could support the wider Endocrine Intelligence programme.
Evaluation required the same discipline. Material used to develop the system could not simply be mixed with the questions intended to measure progress. Otherwise, apparent improvement might reflect prior exposure rather than a genuine change in capability.
These choices later gave new information workflows, Founder Intelligence and clinician-facing product surfaces a consistent base.
The lesson was simple: before the visible product could grow, EndoMind needed a way to keep information in context, distinguish different forms of knowledge and measure change honestly—not merely produce a fluent answer.
Dr Ryizan Nizar MD MRCPUK (Diabetes and Endocrinology)
Previous in the series: Start With the Disease, Not the LLM.