I am an MSc Data Science student studying under Prof Paul Resnick as my program advisor at the University of Michigan, currently residing in Ann Arbor but originally from Detroit, Michigan. After earning a degree in Economics at the same institution, I embarked on a journey through the tech industry, working for Oracle, the supply chain optimization startup LLamasoft, and now as an Operations Analyst at Coupa Software as I complete my MSc.
My research and professional interests lie in the areas of Classical ML, Generative Models, Computer Vision, and Natural Language Processing (NLP). The convergence of these domains drives my passion for exploring innovative solutions and contributing to the field of artificial intelligence.
Born & and raised in Detriot, MI.
Discovered my passion for Mathematics & Statistics.
In mathematical terms, you could say:
\( \cos(\theta_{\rho, \mu}) \approx \cos(\theta_{\rho, \sigma}) \approx 1 \), where
\( \rho = \text{Me} \)
\( \mu = \text{Mathematics} \)
\( \sigma = \text{Statistics} \)
Started College & was inducted into the Honor Society
Transfered to the University of Michigan on a full ride.
Declared Economics as my Major
Finished Undergrad.
Joined Oracle & moved to Boston
Graduated Oracle Academy & recieved promotion
Realized I wanted to return to School
Moved back to Michigan & joined LLamasoft
Began studying for the GRE & learning Python
Accepted into MSc. Business Analytics Program
University of Michigan
Group Research Project was Published
Publication: Simulation Improves Service and Operations at a Franchise Resale Store
Transfered to MSc. Data Science Program
University of Michigan - School of Information
Began Deep Learning Research
Computer Vision, NLP, Generative Models
Finished research project on..
I really appreciate you taking the time to review my portfolio. I hope you enjoyed your time & are inspired to connect with me! Please feel free to download my CV to learn more about my experience, skills, and achievements.
This project showcases a Neural Style Transfer model built on top of the VGG19 architecture, trained to apply the style of a given image to another while preserving the content of the original image.
This project presents a Variational Autoencoder (VAE) trained on the CelebA dataset for 50 epochs. The model can generate novel images of faces, demonstrating impressive results. Two additional features are also provided: a function to display the distribution of the latent space variables and another to manipulate novel images by performing arithmetic in the latent space.
Given that the amount of investments into developing countries to
help accelerate their decoupling journey is significant, our team
wanted to investigate these financial flows and try to determine if
these investments have helped countries lower emissions, all while
still maintaining or growing their economy.
This project is a robust and insightful sentiment analysis tool developed using Python. Its core function is to analyze and visualize the sentiment of news articles associated with climate change. Leveraging the NewsAPI for data retrieval, NLTK for preprocessing, and the transformer library with a pretrained BERT model for sentiment analysis, this project uncovers trends and insights in climate change reporting across various news sources. The visualizations are brought to life using Matplotlib, Plotly, and Wordcloud, offering an engaging representation of the sentiment data.
his project aims to create a classifier that can predict the source of a news article given the content of the article. The sources of news we are considering are CNN, Fox News, MSNBC, and Breitbart. I utilize a pre-trained BERT model to extract features from the articles and perform the classification. After 10 epochs of training, the model achieved an accuracy of 0.9514 with a loss of 0.1718.
Given that the amount of investments into developing countries to
help accelerate their decoupling journey is significant, our team
wanted to investigate these financial flows and try to determine if
these investments have helped countries lower emissions, all while
still maintaining or growing their economy.
This project employs network analysis techniques to map the evolving landscape of PyTorch-related research, identifying key figures and trends in the community through co-authorship networks. It further explores the academic impact of the original PyTorch paper by constructing a multi-level citation network, providing insights into how the paper has influenced and propagated through the research community.
This project presents a Conditional Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP), trained on an animal faces dataset for 50 epochs. The model can generate novel images of different animal faces, although admittedly, there is much room for improvement in this model.
This project presents a Long Short-Term Memory Recurrent Neural Network (LSTM RNN) trained on a large recipe dataset. The model can generate novel, creative recipes, demonstrating the power of RNNs for sequence generation tasks.
This project presents a novel approach to colorize historical grayscale images, employing a UNet Generator and PatchGAN Discriminator. Trained on pairs of historical and modern images, the model can generate colorized versions of historical images of houses around Ann Arbor, Michigan, preserving intricate textures and details.
This project visualizes your hiking journey using the route data exported from Apple Health and metadata from photos taken along the way. The final output is an interactive map with your route, altitude changes color-coded, and photos placed on their respective geolocations.
This project aims to build a personalized game recommendation system for Steam users, employing Neural Collaborative Filtering (NCF) as the underlying algorithm. The model is trained on user-item interactions and leverages playtime as an implicit proxy for user preferences.
This project represents a sophisticated exploration into the realm of Natural Language Processing (NLP), employing Word2Vec embeddings and t-SNE to perform comparative analysis of news articles from different sources.
This project aims to create a classifier that can predict the source of a news article given the content of the article. The sources of news we are considering are CNN, Fox News, MSNBC, and Breitbart. I utilize a pre-trained BERT model to extract features from the articles and perform the classification. After 10 epochs of training, the model achieved an accuracy of 0.9514 with a loss of 0.1718.
This project is a robust sentiment analysis tool in Python, analyzing and visualizing climate change news articles. Using NewsAPI for data retrieval, NLTK for preprocessing, and the transformer library with a pretrained BERT model for sentiment analysis, it uncovers trends in climate change reporting. Visualizations are created with Matplotlib, Plotly, and Wordcloud, providing an engaging view of the sentiment data.
