Start With the Disease, Not the LLM

The more important question

Once the purpose of Endocrine Intelligence was clear, one of the earliest practical questions was which LLM we should use.

The choice mattered, but it was not the most important decision. The more important question was what we were trying to make durable.

LLMs are changing quickly. A strong option today may be overtaken, repriced or delivered differently tomorrow. If the value of EndoMind lived inside one foundation, every major change in technology could force the project to be rebuilt around it.

That would put the emphasis in the wrong place.

Start with the disease

EndoMind is the first clinician-facing product within the broader Endocrine Intelligence programme. Its purpose is to develop useful intelligence in diabetes and endocrinology—not to showcase a particular LLM.

We therefore adopted an early principle: start with the disease, not the LLM.

What should become more valuable

The durable assets are the clinical knowledge structure, specialist teaching, clinical cases, reasoning examples, corrections and evaluation. These are the parts that should become more valuable over time. Each case can reveal a new reasoning challenge. Each correction can expose a pattern that should not be repeated. Each evaluation can make the meaning of improvement more specific.

The selected LLM can still be optimised carefully. The important distinction is between those technology-specific choices and the reusable disease intelligence beneath them.

This has allowed us to experiment through Hugging Face, DeepInfra and Atlas Cloud without making any one of them the identity of EndoMind. Cloudflare can provide durable application infrastructure while the core clinical assets remain portable.

Measuring progress

Starting with the disease also changes how progress is assessed. A general benchmark cannot tell us whether an LLM is becoming genuinely useful for diabetes and endocrinology. We need relevant tasks, clinically meaningful error patterns and specialist review capable of distinguishing a fluent answer from a useful one.

That is why EndoMind is not intended to be a wrapper around a particular LLM. It is the first expression of a broader Endocrine Intelligence approach: build specialist intelligence that can continue to improve even as the underlying technology changes.

The LLM may change. The providers may change. The infrastructure may evolve.

The clinical knowledge, judgement and lessons accumulated around the disease should continue to compound.

Dr Ryizan Nizar MD MRCPUK (Diabetes and Endocrinology)

Previous in the series: The Idea Was to Navigate the Knowledge Universe.