Project / 002
ECG Arrhythmia Classification & TinyML Deployment
A thesis project connecting rigorous ECG model evaluation with constrained on-device inference.
A CNN-based ECG arrhythmia classifier evaluated across patients and deployed for on-device inference on ESP32.
PROJECT / 002
PUBLIC RECORD · REV / 01
01The challenge
ECG classifiers can look convincing without proving that they generalize to patients the model has not seen. The second constraint was practical: carrying the trained model from a research environment onto an ESP32-class device.
02The approach
How the pieces connect
The work uses a CNN with an inter-patient evaluation framework, then converts the trained network with TensorFlow Lite for TinyML deployment. This keeps evaluation discipline and deployment constraints in the same research pipeline.
This record uses verified project scope. Photographs, schematics, measurements, repository links, and test logs will be added only when authentic source material is ready.
03Build scope
Implemented capabilities
04The result
The thesis links model development, inter-patient testing, TensorFlow Lite conversion, and ESP32 inference instead of stopping at an offline accuracy result.
05Technology
- 01Python
- 02TensorFlow Lite
- 03CNN
- 04TinyML
- 05ESP32
06Related work