Project Name

Lorem ipsum dolor sit amet consectetur.

Use this area to describe your project. Lorem ipsum dolor sit amet, consectetur adipisicing elit. Est blanditiis dolorem culpa incidunt minus dignissimos deserunt repellat aperiam quasi sunt officia expedita beatae cupiditate, maiores repudiandae, nostrum, reiciendis facere nemo!

• Client: Threads
• Category: Illustration

Close Project

Project Name

Lorem ipsum dolor sit amet consectetur.

Use this area to describe your project. Lorem ipsum dolor sit amet, consectetur adipisicing elit. Est blanditiis dolorem culpa incidunt minus dignissimos deserunt repellat aperiam quasi sunt officia expedita beatae cupiditate, maiores repudiandae, nostrum, reiciendis facere nemo!

• Client: Threads
• Category: Illustration

Close Project

Project Name

Lorem ipsum dolor sit amet consectetur.

Use this area to describe your project. Lorem ipsum dolor sit amet, consectetur adipisicing elit. Est blanditiis dolorem culpa incidunt minus dignissimos deserunt repellat aperiam quasi sunt officia expedita beatae cupiditate, maiores repudiandae, nostrum, reiciendis facere nemo!

• Client: Threads
• Category: Illustration

Close Project

Recommender System

Recommending songs based on user preferences and item content (think Spotify).

In this project, I explore six recommendation systems using the 1 millions songs dataset: popularity-based, user-user collaborative filtering, item-item collaborative filtering, matrix factorization, cluster-based, and content-based. First, I perform an exploratory data analysis of the 1 million songs dataset, and then train, test, and evaluate the different recommendation algorithms based on prediction accuracy, RMSE, MAE and F1 metrics. I then build and test a hybrid recommendation system based on the best models identified in part I, and apply a weighted joiner function to construct the hybrid recommendations.

• Domain: Data Science
• Category: Recommender system, supervised learning

Link to Project Close Window

Time Series

Forecasting Covid-19 death rates based on vaccination status

The data collected since the start of the novel Covid-19 pandemic offers an excellent opportunity to apply time series forecasting in unique ways. I began by accessing the John Hopkin's Covid-19 world-wide dataset. This dataset is publicaly available on the Johns Hopkin's website and is also updated almost daily on Kaggle.

• Domain: Data Science
• Category: Time series analysis

Link to Project Close Window

Classification

Prediciting customer satisfaction with random forest and xgboost.

This project uses classification and regression methods (Random Forest, XGBoost, XGBoostRegressor) to predict customer satisfaction of customers riding the Shinkansen Bullet Train in Japan. This project was part of my submission for the Great Learning Spring 2023 Hack Linguist Hackathon Competition. The dataset contains 23 customer experience and train/trip attribute features and is pre split into a train and test set. With my solution, I was able to predict customer satisfaction in the test set with >95% accuracy.

• Domain: Data Science
• Category: Classification, supervised learning

Link to Project Close Window

Phylogenomics

Target Enrichment genomic methods combined with gene tree/species tree, gene-flow tests, and cluster analysis.

This project was originally started as my thesis project at Northwestern University and culminated in publication in the Journal of Systematic Biology. It is an example of a fully end-to-end project in all respects. From field collection, to wet lab genomic library construction, to bioinformatics, HPC, data science, dataviz, and publication.

• Domain: Data Science
• Category: Genomics, bioinformatics, phylogenetics

Link to Full Article Close Window

BIS Website

Full stack web application development with Django, Python, CSS, js, and html

I was hired by Buyer Invested Selling to build their company website. I gathered requirements, determined financial and timeline expectations, and worked with the client to deliver a website that met their needs and budget. Like this site, I built the BIS site in a Django framework and deployed it using Github Actions to an EC2 instance on AWS with Route53 DNS service. This allowed me to quickly deploy the site to a cost effective and scalable hosting platform.

• Domain: Software Engineering
• Category: Web Design, full-stack, web app

Link to Site Close Window

Systematic Biology

Target Enrichment and Extensive Population Sampling Help Untangle the Recent, Rapid Radiation of Oenothera Sect. Calylophus

Abstract

Oenothera sect. Calylophus is a North American group of 13 recognized taxa in the evening primrose family (Onagraceae) with an evolutionary history that may include independent origins of bee pollination, edaphic endemism, and permanent translocation heterozygosity. Like other groups that radiated relatively recently and rapidly, taxon boundaries within Oenothera sect. Calylophus have remained challenging to circumscribe. In this study, we used target enrichment, flanking noncoding regions, gene tree/species tree methods, tests for gene flow modified for target-enrichment data, and morphometric analysis to reconstruct phylogenetic hypotheses, evaluate current taxon circumscriptions, and examine character evolution in Oenothera sect. Calylophus . Because sect. Calylophus comprises a clade with a relatively restricted geographic range, we were able to extensively sample across the range of geographic, edaphic, and morphological diversity in the group. We found that the combination of exons and flanking noncoding regions led to improved support for species relationships. We reconstructed potential hybrid origins of some accessions and note that if processes such as hybridization are not taken into account, the number of inferred evolutionary transitions may be artificially inflated. We recovered strong evidence for multiple evolutionary origins of bee pollination from ancestral hawkmoth pollination, edaphic specialization on gypsum, and permanent translocation heterozygosity. This study applies newly emerging techniques alongside dense infraspecific sampling and morphological analyses to effectively reconstruct the recalcitrant history of a rapid radiation.

Link to Full Article Close Window