Welcome, I am a computational and data scientist with specializations in control systems, optimizations, electric power systems, data science, and machine learning. I studied mechanical and aerospace engineering, focusing on control systems and optimization, and their applications to energy systems such as the electric power grid.
After earning my PhD, I researched applications of model-free optimization, unsupervised, and reinforcement learning on the electric power grid, for energy efficiency, grid resiliency, and cyber security. After this, I worked at Hypergiant, developing solutions strategies for mitigating effects of severe weather on the electric power grid. I currently work for Chelle, an AI company leveraging NLP and novel large language models to enable users to chat with their organizations documents.
My skillset includes convex and nonconvex optimization, a bevy of machine learning techniques, and vast experience in neural networks including deep learning and reinforcement learning. An outline can be found at he bottom of this page.
At Chelle, I focus on researching applications natural language processing (NLP) and large language models (LLMs) in creating a platform for users to leverage LLM ability for document resource retrieval. My work includes research into NLP, prompt engineering, and creating interface between the platform and LLMs.
At Hypergiant, I developed solution strategies for mitigating the effects of severe weather on the electric power grid. Here I developed models of weather effects on electricity generation and consumer demand using tensorflow. I implemented a predictive optimization that incorporates the weather models to reduce instances of high line loading. This was all encapsulated in a simulation framework, deployed on a FastAPI endpoint, and incorporating a CICD process over Azure and Docker.
At LBNL, I researched unsupervised and reinforcement learning for optimiztion of energy control algorithms.
I researched extensions of model-free optimization algorithms, and applications to the electric distribution grid.
My dissertation focused on optimal control of two energy systems with nonlinear dynamics: office-scale battery storage (OSBS), and electric distribution grids.
Deep learning is class of machine learning algorithms that uses multiple layers to extract higher-level features from a data set, inspired by the human brain. There is a multitude of comprehensive libraries (e.g. tensorflow) for deep learning.
This project is an OOP approach to creating and training multi-layer neural networks for regression and classification tasks, without the use of those libraries. The neural networks in the project do not use popular libraries (e.g. tensorflow), and are implemented using simple and native python data types, such as numpy arrays. Examples include nonlinear regression, classification of MNIST images, and segmentation of a 2D space according to their L1 or L2 distance from user-chosen points.
Extremum Seeking (ES) is a model-free, online or offline, optimization algorithm that approximates gradient descent. ES estimates the gradient of a system by modulating the input to a system, then filtering and demodulation of an objective function value based on system output.
This project contains three implementations of extremum seeking, meant to be plug-and-play into simulations or real-world applications: 1D-ES, 2D-ES, ND-ES (multi-dimensional). Multiple ES instances of any type can be used in conjunction. Jupyter notebooks contain several simulation examples of the ES implementations on static and dynamic systems.