ESS, Inc.Data Scientist R&D
Oct. 2022 - Mar. 2023Wilsonville, Oregon, United States · On-site• Using multimodal code (Python, SQL, R, and JavaScript) queried and analyzed a large streaming time series based relational database (Java Database Connector) and designed full-stack performance dashboards (Flask, HTML, CSS) to track and visualize key system performance metrics that allowed for the test team to understand the effects of their decisions in specific testing scenarios, thus building feedback loops improving product resiliency. The project succinctly described as ‘completely automated full-stack interactive data visualization product.’
• Developed large-scale deep learning- advanced recurrent neural network (multiheaded attention-based LSTM), model for system optimization and predictive maintenance needs for end use customers. This was made into a data-product in conjunction with site connecting software to be a simple flagging tool for customers to understand their ability to pull power from the energy warehouse, and or, when instead the warehouse needed maintenance.
• Developed key variable analysis using information theory to understand interaction effects among the most important variables making the model as simple and explainable as possible, and most likely to be predictive. This science-based paradigm leverages the uses of multivariate information modeling and reconstructability analysis to statistically determine increased states of probabilistic understanding of common or frequent system states and how to predict the effects of certain relational models on those specific states. This form of data mining is comparable to heuristic based searches, and Bayesian network analysis, yet often much more powerful in its choice of predictive variables and model structures.
• Utilizing partial differential equations, data mining, machine learning (time series specific), and data visualization techniques built replicable, novel scientific algorithmic software for teams within research and development.