This thesis examines the development, validation, and application of a low-cost, reconfigurable wearable sensor system for biomechanical monitoring. The system consists of a 2 degree of freedom angular sensor combined with two 6 degree of freedom inertial measurement units. The system can be reconfigured using 3D printing to develop custom wearable devices for joint specific sensing applications. Here the design is presented and evaluated for validity of angular measures for the knee and wrist. The ultimate purpose of the work is to develop tools and techniques which facilitate longitudinal studies of biomechanical indicators that can be applied in tracking progression and recovery from disease and injury, or to quantify workplace injury risk. To this end, the system was applied first in the knee joint with accompanying gait analysis, and then to the wrist during the performance of a series of occupational tasks. Data produced by the inertial measurement units were then used to train a deep convolutional autoencoder to predict wrist postures which could then be used in existing frameworks for assessment of musculoskeletal injury risk. The data were also clustered and assessed using a novel framework which generates video representations of what participants were doing when the clustered data were produced, allowing a deeper biomechanical understanding of model performance without the explicit need to label the data. This combination of wearable sensing with an unsupervised machine learning pipeline demonstrate a powerful tool for longitudinal studies of human biomechanics in out-of-lab environments.