Cloud to Edge Heart Rate Monitor

A validation device that cross-checks a clinic's heart-monitor readings, split between the cloud and a microcontroller:

A Raspberry Pi 5 plays the patient, replaying MIT-BIH ECG recordings from Google Cloud Storage to a Pico over USB serial
The Pico classifies each heartbeat as normal, supraventricular ectopic or ventricular ectopic using a small neural network I wrote in C (about 115 int8 weights)
A Python exporter writes the trained weights into a C header, so retrained weights go straight into the firmware
The Pico's second core is a watchdog that raises an alarm if the data stops for more than a second, so a stalled feed shows up as an alarm instead of silence

Solo projectDec 2025 – present

Spec

Patient simulator: Raspberry Pi 5
Classifier: Raspberry Pi Pico (RP2040)
CPU: Dual-core Arm Cortex-M0+
Data: MIT-BIH, from Cloud Storage
Link: USB serial, 115,200 baud
Model: Int8 weights, float math
Core 0: Inference engine in C
Core 1: Watchdog, panic strobe after 1 s hang

Why I built it

I built this on my own to explore how edge devices and the cloud can work together. I designed it as a small, independent validation device that could cross-check the readings from a clinic's heart monitor.

The hard part

The hardest part was coordinating the Pico's two cores. Core 0 reads the incoming stream, runs the network and updates a shared timestamp with every sample. Core 1 waits for the first sample, then monitors that timestamp and raises an alarm if it stops updating for more than a second. Both cores share one LED, so the output had to stay distinguishable from the alarm: one to three blinks for the beat type, and a fast 30 ms strobe for a fault.

Code on GitHub

Next projectLonghorn Neurotech →