Plain Language Overview¶
This page explains the SDK in simple terms.
What Is IINTS-AF?¶
IINTS-AF is a safe testing environment for insulin algorithms.
You can think of it as a flight simulator: - You test ideas in simulation. - You measure what happens. - You keep an audit trail. - You do not treat real patients with this SDK.
What Happens During a Run?¶
When you run a simulation, the SDK does this:
- Reads glucose and scenario events (meal, exercise, sensor issue).
- Checks if glucose input is biologically plausible (
InputValidator). - Gets an insulin suggestion from your algorithm.
- Optionally reads AI forecast signals (advisory only).
- Applies hard deterministic safety checks (
IndependentSupervisor). - Simulates the patient response and stores all outputs.
Why “Open Logic” Matters¶
IINTS-AF is not “AI decides everything.”
- Layer 1:
InputValidatorfilters impossible sensor values. - Layer 2:
Predictorestimates future glucose (optional, advisory). - Layer 3:
IndependentSupervisorcan block or reduce unsafe doses.
Final dosing is always safety-constrained.
Who Should Use It?¶
- Researchers testing control algorithms.
- Developers building simulation pipelines.
- Clinical innovation teams preparing pre-clinical evidence.
- Students learning diabetes algorithm validation.
What It Is Not¶
- Not a medical device.
- Not cleared for direct patient treatment.
- Not clinical decision support in production care.
5-Minute Start¶
pip install iints-sdk-python35
iints quickstart --project-name iints_quickstart
cd iints_quickstart
iints presets run --name baseline_t1d --algo algorithms/example_algorithm.py
You will get output files like:
- results.csv (time-series outputs)
- audit/ (decision trace)
- report.pdf (visual summary)
“I Need X” Quick Map¶
- “I want a first run” ->
README.md - “I want full CLI commands” ->
docs/TECHNICAL_README.md - “I want full architecture details” ->
docs/COMPREHENSIVE_GUIDE.md - “I want the real research sources behind assumptions” ->
docs/EVIDENCE_BASE.md - “I want predictor training” ->
research/README.md - “I want step-by-step notebooks” ->
examples/notebooks/README.md
Short Glossary¶
- TIR: Time in Range (usually 70-180 mg/dL).
- IOB: Insulin On Board (active insulin still working).
- COB: Carbs On Board (carbohydrates still absorbing).
- Fail-soft: Keep simulation running using last safe value when input is invalid.
- Dual-Guard: Predictor + deterministic supervisor, with safety as final authority.