Case Study · Medical Device Software

Submission-ready software for a connected glucose monitor

About the project

A couple sitting together on a sofa at home, looking at a health app on a smartphone

The client

A health-and-wellness company with a shipping app and a loyal user base, moving into regulated territory with a connected continuous glucose monitor.

The challenge

Glucose is a metric a user may dose or eat against, so the software had to meet a standard that survives regulatory review, inside an app never architected as a regulated system, and without slowing the consumer side down.

A tablet and smartphone displaying glucose analytics dashboards on a soft lavender background

The approach

Treat the glucose capability as a bounded, regulated module with an auditable interface to the existing app, so measurement is controlled while the consumer experience stays fast.

The impact

A glucose feature that feels native to an app users already trust, with a documentation trail built to stand up to review and a clean data lineage from sensor to store.

The solution

The product serves multiple roles (the user, care circle, and clinicians), so the solution required native iOS and Android apps and a web view over one bounded, validated glucose module.

Glucose trends and alerts screen with a predictive heads-up card warning that low glucose is likely soon, above a twelve-hour trend curve

Engagement · AI Velocity Pod

Standards · IEC 62304 + ISO 14971

Step 1

Classification & risk planning

Define intended use and users
Classify as SaMD and set the IEC 62304 software safety class
Open the ISO 14971 risk file and risk management plan
Map the target regulatory route

Step 2

Engineering & integration

Bounded regulated module for all glucose measurement
BLE data path with integrity checks and safe OTA firmware
Native integration into the existing app via a controlled interface
Encrypted, auditable telemetry from sensor to store

Step 3

Verification & submission

Multi-level V&V with a requirement-to-test matrix
Continuous risk file, hazards traced to controls
Design history file and connected-device cybersecurity docs
Submission package for regulatory review

AI development

The reading itself stays a faithful measurement. The intelligence sits around it, and every model that touches a measurement or an alert is treated as part of the regulated system.

01Predictive alerting

02Adaptive drift compensation

03Alert personalization

04Event-driven suggestions

Predictive alerting reads the recent glucose trajectory to warn of a probable low or high before it arrives. Drift compensation keeps accuracy stable across the full sensor wear period. Personalization tunes thresholds and quiet hours to the individual to cut alert fatigue.

Models are validated on representative data with performance bounded and documented. Locked algorithms are used where the submission needs fixed, provable behavior, and anything meant to improve over time runs through a predetermined change control plan, never a silent update.

Kept outside the boundary

Any AI on the engagement side (coaching prompts, educational content) stays outside the regulated system and outside the regulated claims.

A nurse applying a continuous glucose monitor sensor to a patient's upper arm in a bright clinic

The human stakes

Behind every glucose reading is a decision someone makes about their health. That is the standard the software is held to.

Tech stack

Device & connectivity

Bluetooth Low EnergyOTA firmwareIR PPGRed-LED opticalNon-contact temp sensor

Mobile

iOS 14+Android 6+native / Flutter (confirm)

Backend

.NET / Java / Python / Node / Go (confirm)

Cloud · HIPAA-eligible

AWS / Azure / GCP (confirm)

Data & AI

Real-time streamingOn-device MLCloud MLData lake

Security & records

End-to-end encryptionAudit logging21 CFR Part 11

Submission-ready by design

Evidence produced alongside the code, not reconstructed before a deadline. Accuracy is validated against a reference method and read through a consensus error grid, so the clinically dangerous errors are what the device is held to.

IEC 62304ISO 14971ISO 1348521 CFR Part 11FDA / EU MDRISO 27001ISO 9001SOC 2 Type IIHIPAA-ready

Proof of Scale

Before we built our AI factory, we architected the platforms for some of the fastest-growing enterprises...

Assurecare

Connected Care Platform for 53M+ Members

Ailoitte helped power AssureCare’s patient-centered platform that connects payors, providers, and pharmacies to improve access, reduce cost, and strengthen care coordination.

Read Case Study