A Rapid Pipeline for Training and Deploying ML Models on WeBe Band

Edwin Kay1, Asmita Asmita1, Houman Homayoun1, Mahdi Eslamimehr2
1UC Davis, 2Quandary Peak Research


Abstract

This paper presents an automated, end-to-end TinyML framework that streamlines the training, deployment, and on-device evaluation of machine-learning models on WeBe Band, a research-grade, healthcare-focused wearable device developed by Health-eTile. The WeBe Band integrates multimodal physiological and motion sensors and provides a practical platform for evaluating edge AI under realistic wearable constraints. By abstracting embedded firmware complexity and enabling rapid over-the-air deployment and profiling, the proposed system allows researchers to efficiently explore trade-offs between model complexity, inference latency, and memory usage on microcontroller-class hardware. Experimental results show that lightweight classical models achieve reliable real-time performance with minimal resource overhead, highlighting the effectiveness of the framework for wearable edge AI applications.