An IoT Research Tool for Data Acquisition, Annotation, and ML Deployment on Custom Physiological Wearables

Modern biomedical research on novel wearable sensors requires flexible tools that can integrate heterogeneous sensors, acquire multi-modal physiological signals, and enable streamlined experimentation with machine learning (ML) pipelines. However, a practical end-to-end toolchain supporting acquisition, dataset curation, and iterative model deployment for such emerging signals remains limited. Recent advancements in biocompatible Aluminum Nitride (AlN) piezoelectric wearables have demonstrated unobtrusive monitoring of cardio-respiratory and mechano-acoustic signals (e.g., pulse waves and heart sounds), typically validated against electrocardiogram (ECG) signals, highlighting the need for structured data management and annotation workflows. This paper presents a modular edge–cloud software framework designed for the rapid integration and evaluation of custom Bluetooth-enabled physiological devices. The framework provides an extensible acquisition interface for device registration and configuration, near-real-time visualization, and event-centered dataset generation with annotation and versioning capabilities. The system supports iterative ML development by enabling dataset export for training and deployment of lightweight inference models on an edge gateway, allowing real-time operation without persistent cloud connectivity. The platform is demonstrated using a prototype device that acquires ECG, AlN piezoelectric signals (cardiac/respiratory mechanical activity), body temperature, and inertial motion data, enabling end-to-end validation of acquisition, event generation, annotation, and edge inference within a single workflow. Overall, this framework accelerates development cycles and facilitates the exploration of new wearable-sensing solutions in biomedical research.