Accessible Discovery

What exactly is a "Smart" Insulin Pump?

Demystifying the algorithms that drive autonomous care.

IINTS-AF SDK Logo

Understanding the "Recipe"

An algorithm is simply a recipe a set of rules. In a standard pump, you set the rules (e.g., "delivered 2 units for this meal"). In a smart pump, the computer attempts to determine the rules itself based on your glucose levels.

The "Black Box" Problem

Major manufacturers often keep these algorithms hidden. While you see the outcome the insulin being delivered you often don't understand why the decision was made.

  • Hidden Logic: You lack insight into the actual decision-making process.
  • Trial & Risk: You cannot safely experiment with how the system reacts to stress or exercise.

IINTS-AF: The Research SDK

IINTS-AF has evolved from a standalone experiment into a comprehensive Software Development Kit (SDK). It provides an independent research framework that allows developers and academics to build, test, and validate AI algorithms within a safe, deterministic environment.

Data Registry

Global Data Access & Integration

Before an algorithm ever controls a real pump, it must be validated against diverse, real-world physiological data. The IINTS-AF SDK features a built-in Data Registry, allowing researchers to instantly fetch and benchmark against world-class datasets, including the Jaeb Center AIDE T1D, Tidepool Big Data Donation, and the OpenAPS Data Commons. We provide the bridge between raw clinical data and actionable AI safety.

"Simulation is the first step toward safety."
Prediction Point

Ohio T1DM Physiological Sandbox

Live System Logic

The algorithm identifies trends in the glucose curve, neutralizing peaks before they occur.

Clinical Transparency

From "Black Box" to "Open Logic"

Dual-Guard Security Architecture

We do not believe in blind trust in AI. That is why the IINTS-AF SDK introduces a layered defense strategy:

Layer 1: Input Validator – A biological plausibility filter that scrutinizes sensor data. It prevents algorithm "hallucinations" caused by sensor artifacts or noise.

Layer 2: Independent Supervisor – A deterministic safety layer that audits every AI decision against hard medical rules (IOB capping and hypoglycemia prevention) before any action is taken.

01

Validation

The InputValidator filters noise and artifacts from the Ohio T1DM CGM data, ensuring only physiologically plausible values reach the controller.

02

Intelligence

An LSTM Neural Network identifies hidden insulin sensitivity patterns to predict future glucose trends.

03

Supervision

The Independent Supervisor validates the dose in microseconds on the Jetson Nano, ensuring safety logic always overrides AI suggestions when risks are detected.

"Code shouldn't be a secret when it's managing a life."

Research Access

Take Control of Your Technology

IINTS-AF is a fully Open Source research platform designed to provide transparency into autonomous care. We believe you should be the master of your own technology through understanding.

Simulation

High-fidelity in-silico testing using the Ohio T1DM physiological dataset.

Edge AI

Optimized for the NVIDIA Jetson ecosystem. Run high-speed inference and 24/7 safety monitoring on a dedicated hardware 'waakhond' that ensures your SDK remains bug-free and responsive.

Open Access

DUMMY

MIT Licensed framework for independent medical tech research and study.

Join the Research

The IINTS-AF SDK is now a complete ecosystem for transparent, explainable, and safe autonomous care. Download the latest version on PyPI or explore the source code on GitHub.

Explore on GitHub
SDK Experience

Built for Developers & Researchers

Getting started with medical AI research shouldn't take weeks. With IINTS-AF, it takes seconds.

One-Command Install

pip install iints-sdk-python35

Instantly set up the complete environment including safety engines and simulators.

Instant Data

iints data fetch aide_t1d

Built-in connectors for the world's leading T1D datasets.

Deterministic Testing

iints run --stress-test

Run thousands of simulations to find edge-case bugs before they reach a patient.

"IINTS-AF represents a shift toward explainable autonomous care, where human safety logic serves as the final authority over artificial intelligence."

Technical Architecture

Component Specification Status
SDK Version v0.1.10 (Global Registry Update) Stable Release
Safety Engine Independent Supervisor (Deterministic) Active
Data Engine Multi-Dataset CLI & Registry New Feature
Edge Latency 0.23 µs (Supervisory Overhead) Benchmarked
Edge Hardware NVIDIA Jetson Optimized Active