This paper presents a rapid pipeline that enables training, optimization, and deployment of machine‑learning models directly on the WeBe Band wearable. The system automatically generates hardware‑efficient models, offering AutoML, hardware‑aware quantization, and performance profiling to respect latency, memory, and power limits. The main stages are data preprocessing, model search, quantized compilation, and OTA update. By tightly integrating the open‑source Piccolo AI ecosystem with an automated toolchain, we produce firmware artifacts that can be flashed into the WeBe core. The pipeline supports both classical algorithms and lightweight neural networks and includes on‑device tools to measure inference latency and memory footprint. Experiments show that classical models achieve millisecond‑level response on a microcontroller, while lightweight networks require careful resource management. The focus is on system‑level automation rather than novel architectures, helping researchers iterate quickly and evaluate models on target hardware. Although demonstrated on the WeBe Band, the workflow is designed to be portable to other edge devices. Review