Learning Path

This learning path is the recommended route for someone who did not build IINTS-AF and wants to understand it independently.

You do not need diabetes-device experience to begin. Complete the modules in order; each module has a practical checkpoint before you continue.

Module 1: Understand The Scope

Goal: know what IINTS-AF is, and what it is not.

Read:

1. Plain-Language Overview
2. Project Boundaries
3. Core Concepts

Checkpoint:

• You can explain the difference between a virtual patient and a real patient.
• You know that candidate algorithms are experimental.
• You know that local AI may explain results but does not calculate authoritative physiology or dosing.

Module 2: Install And Verify

Goal: create an isolated Python environment and verify the SDK.

Read Installation, then run:

python3 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install "iints-sdk-python35[full,mdmp]"
iints doctor --smoke-run

Checkpoint:

• iints --version prints a version.
• iints doctor --smoke-run completes without a blocking error.
• You know which virtual environment contains the SDK.

Module 3: Run One Experiment

Goal: create a deterministic run bundle.

Follow First Run:

iints demo quick --output-dir results/first_run

Checkpoint:

• results/first_run/results.csv exists.
• A report and manifest are present in the run folder.
• Repeating the same command with the same seed gives reproducible results.

Module 4: Read The Evidence

Goal: understand what the run produced before changing an algorithm.

Read Understand A Run. Inspect:

1. the time-series CSV
2. the report
3. the run metadata
4. the manifest and audit information

Checkpoint:

• You can identify glucose, carbohydrate, insulin, and safety-event columns.
• You can distinguish a plotted trajectory from proof of clinical realism.
• You can find the seed, SDK version, and run settings.

Module 5: Build A Complete Workflow

Goal: move from a demo to an explicit project with controlled inputs.

Follow Complete First Workflow. You will:

1. scaffold a project
2. select an algorithm, patient, and scenario
3. run a simulation
4. validate the output
5. review the evidence bundle

Checkpoint:

• The patient, scenario, algorithm, duration, time step, and seed are recorded.
• The output can be traced back to those inputs.
• Any warnings are documented rather than hidden.

Module 6: Choose A Specialisation

Continue only with the route you need:

Route	Start with	Main question
Simulation studies	Scientific Workflow	How do I compare algorithms reproducibly?
Data quality	Certification Quickstart	Is this dataset sufficiently documented and valid for the intended analysis?
Local AI	AI Assistant	How can a local model explain already-computed evidence safely?
Glucose forecasting	Glucose Forecast Model	How do I train and evaluate a research predictor without data leakage?
Physiology	Physiology Reference	Which equations, parameters, assumptions, and limitations are implemented?
Hardware	Hardware Hub	How do I run a bench-only hardware experiment?
SDK development	Developer Portal	How do I change the code without breaking contracts or reproducibility?

Research Habits To Keep

• Fix the protocol before looking at the final result.
• Preserve seeds, versions, configuration files, and manifests.
• Use subject-level splits for patient datasets whenever possible.
• Compare against simple baselines, not only against your previous model.
• Report failed runs and excluded data.
• Treat simulation realism, predictive accuracy, and clinical validity as separate claims.
• Keep private or licensed datasets outside Git.

The Scientific Workflow expands these rules into a full study process.