Canara (now CPG Data Center Innovators) is a major company in the smart battery sector, successfully applying predictive technology to batteries and deploying this across their services. They monitor hundreds of power supply battery units in data centers, and forecasts the failure time so that their customers can swap their batteries out well before failure. The better the forecast, the more money their customers can save by swapping their batteries later. Canara was using a crude polynomial fit to perform the forecast before we added a Kalman filter forecast, which alone improved the prediction relative to the polynomial fit and on top of that, built a machine learning model that improved the forecast further. A large portion of the project involved handling the multitude of battery makes and models and cleaning up/smoothing/aggregating the raw, noisy time series data (which was of different collection frequencies and calibrations)