Glucose Trend Alert System

Overview

Why this project matters

This project connects software engineering, healthcare data analysis, validation testing, and medical-device software awareness in a way that directly supports my long-term interest in diabetes care technology and human-centered medical systems.

Problem Space

Glucose behavior changes over time

CGM-style systems are not only about a single glucose reading. Direction, rate of change, risk-zone context, and persistence can all affect how useful an alert feels to a user.

System Goal

Turn readings into interpretable signals

The system processes synthetic readings into risk zones, smoothed glucose values, rate-of-change calculations, trend labels, simulated alerts, and exportable reports.

Portfolio Value

Healthcare software thinking

The project shows how requirements, modular design, validation, dashboarding, and user-centered alert behavior can come together in a healthcare-inspired system.

Important disclaimer: This project uses synthetic data only. It is not a medical device, is not intended for clinical use, is not medical advice, is not a diabetes management tool, is not connected to a real CGM, does not use real patient data, and is not affiliated with Abbott, FreeStyle Libre, Dexcom, or any medical device company.

System Architecture

A modular CGM-style analytics pipeline

The project is structured as a modular pipeline so each stage can be tested, documented, and extended independently.

01

Synthetic Data Generator

02

Risk Zone Classifier

03

Trend Detector

04

Alert Engine

05

Reports + Dashboard

Inputs

Synthetic CGM-style readings

The data generator creates simulated glucose readings at 5-minute intervals across baseline scenarios, including stable glucose, rising glucose, falling glucose, post-meal spike, and overnight drop.

Stable Glucose Rising Glucose Falling Glucose Post-Meal Spike Overnight Drop

Outputs

Processed analytics and reports

Pipeline outputs include processed CGM CSV data, risk-zone summaries, trend summaries, alert summaries, and dashboard-ready data for Streamlit visualizations.

Processed CSV Risk Summary Trend Summary Alert Summary Dashboard

Implementation

What the system does

Data Generation

Generates synthetic CGM-style readings and controlled glucose scenarios without using real patient data.

Risk Zones

Classifies simulated glucose values into illustrative risk zones such as very low, low, in range, high, and very high. These zones are for simulation only and are not personalized medical guidance.

Trend Detection

Calculates glucose rate of change in mg/dL per minute and labels readings as rapidly rising, rising, stable, falling, or rapidly falling.

Alert Engine

Generates simulated alerts with alert type classification, priority assignment, user-centered messages, persistence logic, cooldown logic, and suppression reason tracking.

Dashboard

Visualizes synthetic glucose traces, risk-zone bands, smoothed trend lines, latest metrics, active simulated alerts, processed data tables, and CSV download options in Streamlit.

Validation + Documentation

Built with verification and traceability in mind

The project includes automated tests and medical-device-inspired documentation to demonstrate requirements-driven thinking, risk-aware design, and responsible non-clinical-use boundaries.

Automated Testing

pytest validation

Tests cover synthetic data generation, risk-zone classification, trend detection, rate-of-change calculation, alert generation, persistence logic, cooldown logic, and end-to-end pipeline behavior.

Documentation

Medical-device-inspired artifacts

Documentation includes software requirements, a software design document, a risk management summary, a validation plan, a requirements traceability matrix, and a regulatory/clinical use disclaimer.

User-Centered Design

Alert fatigue awareness

Persistence and cooldown logic were included to explore the challenge of making alerts meaningful without overwhelming users with repeated or noisy notifications.

Core trend calculation concept

rate_of_change = (current_glucose - previous_glucose) / minutes_elapsed

Trend labels:
  >= +2.0 mg/dL/min   Rapidly Rising
  +1.0 to +1.99       Rising
  -0.99 to +0.99      Stable
  -1.0 to -1.99       Falling
  <= -2.0 mg/dL/min   Rapidly Falling

Medical Device Relevance

Safety-aware, not compliance-claiming

This project does not claim regulatory compliance. Instead, it shows that I understand how healthcare software needs clear intended-use limits, validation, documentation, traceability, risk-aware logic, and user-centered design.

What it demonstrates

Responsible healthcare software framing

The project demonstrates intended-use boundaries, synthetic data handling, requirements-driven development, verification and validation testing, risk-aware alert design, human factors awareness, and clear non-clinical-use disclaimers.

Why it matters

Relevant to diabetes technology

Diabetes care systems sit at the intersection of sensors, software, analytics, alerts, user behavior, and safety-critical communication. This project is a portfolio-scale exploration of that intersection.

Compliance note: Standards such as ISO 14971, IEC 62304, FDA design controls, and 21 CFR Part 11 are not implemented as compliance requirements in this project. They are best understood here as future learning directions for deeper medical-device software development.

Technical Stack

Tools and technologies

Core Stack

Python analytics pipeline

Built with Python, pandas, NumPy, Plotly, Streamlit, pytest, setuptools, pyproject.toml, Markdown documentation, Mermaid diagrams, Git, and GitHub.

Python pandas NumPy Plotly Streamlit pytest Mermaid GitHub

Skills Demonstrated

Healthcare analytics and software engineering

Demonstrates synthetic data generation, modular pipeline design, data processing, alert logic, automated testing, dashboard development, report export, and risk-aware engineering.

Synthetic Data Risk Classification Trend Detection Alert Logic Dashboarding Validation Testing

Future Direction

How I would extend it

Backend

FastAPI + database logging

Add a FastAPI backend, SQLite or PostgreSQL logging, and structured API endpoints for simulated ingestion and report access.

Modeling

Predictive risk modeling

Add a scikit-learn risk prediction model, configurable scenario builder, and synthetic sensor dropout/noise validation.

Systems

Embedded-style simulation layer

Extend the project toward simulated edge-device ingestion, local processing, and communication between embedded and cloud-style components.

View GitHub Repository Return to Portfolio