The objective of this research is to create an end-to-end deep learning pipeline for on-device real-time Sesimocardiogram (SCG) Signal Quality Indexing (SQI) and assessment. SCG is a measurement of the vibrations of the chest wall in response to the heartbeat, and can be measured with non-invasive wearable devices. The first aim established a rigorously labeled SCG dataset to serve as the foundation for algorithm development. Using a custom open-source graphical user interface, SCG waveforms from a porcine hypovolemia protocol were expertly annotated for signal quality and cardiac timing intervals. These annotations demonstrated strong correlation with gold-standard catheter measurements that were simultaneously obtained, providing the biomedical community with a technically validated dataset to accelerate SCG-based real-time cardiac monitoring research and enhance reproducibility. The second aim leveraged this curated dataset to develop a deep learning framework for automated SCG signal quality assessment. On porcine data, these temporal deep learning models outperformed traditional Dynamic Time Feature Matching (DTFM) baselines by retaining significantly more usable beats and reducing timing estimation variance. Utilizing a few-shot learning paradigm, the framework demonstrated robust cross-species transfer to human subjects, achieving high accuracy with only seconds of patient-specific calibration. The third proposed aim bridges the gap to clinical translation by optimizing these cross-species SCG models for on-device inference. By deploying the SQI framework directly onto resource-constrained wearables, this work will enable the real-time classification of high- and low-quality SCG signals at the edge, ultimately facilitating continuous, scalable, and robust cardiomechanical monitoring in ambulatory settings, as well as physiological closed-loop control based on these models operating in real-time.