This past summer, I interned at Google where I developed time series prediction and computer vision object detection machine learning models with quantitative analysts and machine learning engineers for Google Play Commerce ML using Python, TensorFlow, and statistical methods to forecast and improve sales.
In the spring of 2018, I interned at Amazon, specifically, AWS Deep Learning, where I worked with applied scientists to implement computer vision and NLP models using MXNet and TensorFlow. I also contributed large scale language models to the Apache MXNet deep learning framework.
I interned at the Georgia Tech Research Institute in the summer and fall of 2017 where I developed a machine learning platform called EMADE (Evolutionary Multi-Objective Algorithm Design Engine) using genetic programming to automatically create combinations of machine learning algorithms (e.g., deep neural networks, gradient boosted decision trees, SVMs, and clustering algorithms) that achieve better performance than any individual algorithm on multiple objectives (e.g., F1 score, mean squared error, and complexity of GP tree).
I'm continuing what I've learned in my GTRI internship through the Automated Algorithm Design Vertically Integrated Project where I'm applying the EMADE to Kaggle data science competitions and continuing to improve the framework. Under Dr. Charles Isbell and Himanshu Sahni, I did research on AI decision making through hierarchical deep reinforcement learning with state space decomposition, action space hierarchy, and object-centric attention mechanism.
"Computer science is no more about computers than astronomy is about telescopes."
- Edsger Dijkstra
Created video game AI using Asynchronous Advantage Actor-Critic (A3C) model and Deep Q Networks for OpenAI Gym reinforcement learning environments with Python and TensorFlow for neural networks. Only done with CartPole problem but hope to scale to more complex problems (like Atari) when I get a stronger computer.
Competed in Kaggle data science competitions with EMADE machine learning framework developed at GTRI for house price prediction and image recognition competitions.
Winner of HackDuke Best Use of Facebook API prize! Personal carbon footprint tracker that enables the ordinary individual to take action against climate change. Android app in which I wrote much of core Java that integrated the various XML layouts and scoring algorithms. I also made the app interactive. Click here for Devpost link. Also, click here for GitHub link.
Developed Android application to track the quality of water through crowdsourced information and used Google Maps API. Exceeded expectations for project with Firebase to store user data and additional user tracking features. Also created a web app with Express and Node to accompany app for more experience.
HackGT project that helps potential job applicants form better relations with companies, creating richer dialogue between recruiters and individuals. Website created using Indeed's API, Wikipedia's API, Python and Flask. First hackathon project but somewhat slow. Click image for link.