Pulse Check.

Capstone Project

Our Objectives

Our project addresses a gap in accessible cardiac monitoring by developing a mobile, edge-based arrhythmia detection system. Pulse Check ingests ECG data that users already generate through everyday wearables and performs on-device classification to surface early signs of irregular rhythms. By keeping analysis local to the device, the system delivers faster results, stronger privacy, and a more proactive approach to heart-health monitoring.

We proved that consumer-grade ECG data paired with signal processing and on-device ML can reliably flag early signs of arrhythmias.

Exploring the Science

Our model and application is built on the idea that everyday ECGs contain more information than consumer devices currently use. Our work uncovers that hidden signal through a combination of physiology, signal processing, and on-device machine learning.

We begin by preprocessing the raw ECG waveform: applying a Butterworth band-pass filter to remove low-frequency drift and high-frequency noise, downsampling to 500 Hz to standardize recordings to match our training data, and finally z-score normalizing the signal so the model focuses on shape and rhythm rather than absolute amplitude. These normalized traces are then passed into our neural network trained to distinguish normal sinus rhythm from arrhythmias.

The model looks for irregular timing, abnormal beats, and structural deviations in the waveform that often precede more dangerous conditions

.

The result is a fast, private, edge-based system that brings clinically relevant analysis directly to the user.

Key Hypotheses

Consumer-grade ECGs can detect more than AFib when paired with proper signal processing
Lightweight neural networks can run efficiently on a phone and still achieve clinically meaningful accuracy
On-device inference would improve speed, privacy, and user trust
Proactive feedback will increase likelihood that users seek timely clinical care

References

Demo

Previous Project

Next Project