The objective is simple to state but difficult to build: a system that learns how a specific metabolism behaves in practice.
From recording to understanding
Today, most glucose systems are recorders. They collect data, display graphs and sometimes apply simple predictive models. They show glucose values, trends and basic predictions based on fixed assumptions such as static insulin sensitivity or standard absorption models. These tools are useful, but they do not truly understand how a specific body behaves.
In reality, physiological response varies. Insulin action changes across time, meals behave differently depending on composition and context, and repeated patterns often produce consistent but personalised outcomes. Static models cannot fully capture this.
What is missing is a system that learns continuously from real historical data and adapts its internal understanding over time.
The goal of islet and glucore is to move beyond recording into understanding. The long-term objective is a local-first metabolic intelligence system that learns, over time, how a particular body responds in practice. Not in theory or from generic population models, but from real historical behaviour.
This means learning patterns such as:
Rather than relying on fixed parameters, the system aims to continuously refine its internal model using observed data.
The purpose is not automation, but clarity: carefully constrained decision support grounded in what has actually happened before.
Why local-first
Health data is long-lived, sensitive and operationally important. Systems that depend on external cloud services introduce additional points of failure, opacity and loss of control.
A local-first approach keeps core operation under direct control. Data remains accessible even without network connectivity. Behaviour is observable. Failure modes are predictable. The system can be audited and understood rather than treated as a black box.
This does not mean isolation from the wider ecosystem, but it does mean that the critical path remains self-contained and reliable.
islet is designed to remain local-first and self-hosted. Control, reliability, observability and safe failure behaviour are treated as core requirements rather than secondary concerns. The system should fail safely, explain its reasoning and make uncertainty visible rather than hidden.
What makes this difficult
Building a system that genuinely learns metabolic behaviour is not straightforward.
Glucose data is noisy and sometimes wrong. Sensors lag. External context such as illness, stress, exercise and sleep affect outcomes but are difficult to measure precisely. Physiological response is variable rather than deterministic.
Any useful model must operate under uncertainty. It must tolerate incomplete data, avoid false confidence and fail safely when inputs are unreliable. Accuracy alone is not enough. The system must also be transparent, explainable and bounded by explicit safety constraints.
The role of glucore
glucore is the modelling and analysis engine within the system. Its role is to transform raw data into structured understanding.
Rather than simply predicting future glucose values, it aims to:
The intention is not to produce opaque predictions, but to build a progressively improving and auditable representation of metabolic behaviour.
What success looks like
Success is not a perfect prediction curve or a fully automated system.
A successful outcome would be a system that reflects how a specific body actually behaves, with predictions that are transparent, bounded and explainable. Uncertainty would be clearly communicated rather than hidden. Behaviour would remain reliable under imperfect data. Decision support would be grounded in real historical outcomes and the platform would improve over time rather than remaining static.
In practical terms, success means clearer understanding, fewer surprises and better-informed decisions.
The system does not replace clinical judgement and does not make autonomous dosing decisions. Its purpose is to support understanding, not to remove responsibility.
The direction is long-term and iterative. The foundation is being built first: reliable ingestion, correct storage, safe modelling and observable behaviour. The intelligence layer develops on top of that, gradually, through measured refinement rather than assumption.
The near-term work is building a trustworthy foundation: ingestion, storage, validation and modelling that can be measured against reality.
Safety note: islet is a personal project and is not medical advice. It does not make dosing decisions.