Description: Year long project to assess environmental and socioeconomic impacts of the COVID-19 pandemic on Texas. I worked in a team of 8 students of various majors to design and implement a data science pipeline to assess COVID's impacts. This pipeline collected the most recent environmental and socioeconomic county level data, compared historical and present data using statistics and time-series modelling, and visualized impacts by metropolitan area and county with a dashboard.

To implement this pipeline, I first utilized Selenium to automate downloading historical data from the Texas Comission for Environmental Quality (TCEQ) due to their lack of an API. Then, I leveraged Python's Pandas and Statsmodels packages to group data by county and metropolitan area and model what indicator levels would have been if the pandemic never occurred. Finally, to improve interpretability and simplify analysis of the data, I created an interactive dashboard using Dash, a Python dashboarding package.

The results of this research and the subsequent dashboard were presented at the American Geophysical Union conference. Furthermore, the paper is in the process of being published.

Packages used: Dash, Plotly, Matplotlib, Statsmodels, Pandas, Selenium.

Description: Modeled demand for Uber rides using SARIMA + Facebook's Prophet models. Found Prophet to be more effective due to its ability to anticipate irregular peaks in demand for rides.

Packages used: Pandas, Scikit-learn, Prophet, SARIMA

Description: RiceBook is a social media app I created while in COMP 431: Web Development. To create this app, I had to master the frontend and backend by using a MERN stack (MongoDB, Express, React, and Node.js).

Packages used: MongoDB, Express, React, Node.js, HTML, CSS, Bootstrap

Description: StudyFlow is reading tool I created with a team of two as part of HackRice 2021. StudyFlow seeks to enhance reading with functionality to summarize text, define words, and elaborate on topics the user selects. My role in creating this app was to design and implement the frontend interface. My goal in designing StudyFlow's UI was to create a UI that was unobtrusive, allowing the user to focus on reading first and foremost. This meant placing the text center frame and the reading assist features to the sides of the text. To implement this design, I utilized HTML, Bootstrap, CSS, and JavaScript.

In the end, our project did not end up winning any awards; however, I still think the experience was worth it. To complete this project, our team ended up pulling an all-nighter. During this time, our team had fun working together and learning from one another. I learned a lot about designing a backend from our team member who implemented the API to define, summarize, and elaborate upon text. This knowledge would help me later in COMP 431: Web Development, when I created the backend for my social media web app RiceBook.

Unfortunately, only the summarize function works as of December 2021. The backend utilizes a webscraper on a Google search that now fails due to a change in how Google returns search queries. To avoid this downfall in the future, I'd build the backend only using APIs instead of using web scraping, which is unreliable due to how often web pages are updated/changed.

Technology used: JavaScript, CSS, Bootstrap, HTML5.

Description: Month long project to design a Java based, peer-to-peer architecture chatting app. Project began with a two-week design phase to design an API capable of handling network transmission of all types of messages that could be sent between users. I worked in a group of 14 to design an API that we presented to the class.

After the design phase, I worked with a partner to implement our ChatApp. We designed our ChatApp to be made up of two units: a main MVC that runs on the client's computer and mini-MVCs that are created inside the main, which represent chatrooms the user is a member of. I implemented the Main MVC, which is responsible for connecting remote clients together and creating chat rooms for multiple users to communicate with.

I found this project to be powerful in that it embodied the entire software development process. Rather than jumping straight into code, we spent time designing a quality API and architecture. This was pivotal in writing code that was modular and semantically divided, making it easy to understand and expand later if required.

Packages used: Java RMI, Java GUI components

Description: In this project, the goal was to implement PageRank, an algorithm well-known as the basis of Google's original search engine. The algorithm functions on a directed graph of web pages, where an edge exists between webpages A and B if there exists a link to page B on page A. On execution, the algortihm simulates a web surfer who randomly selects a link on the current page to navigate to next. The PageRank is the probability that when the web surfer stops clicking, the user will be on a specific page.

I implemented this algorithm using SQL. As this algorithm is iterative and SQL is a language that performs best with declarative code, I did as much calculation as possible declaratively. This project was critical in improving my SQL skills and my understanding of data operations I often perform in Pandas that originated in SQL (Ex: Join).

Description: In this assignment, I had to predict whether a flight would experience a delay of greater than an hour based predictor variables such as the weather, distance traveled, and departure time. Utilizing sklearn, I built three models to do this task:

• Logistic regression with an L2 penalty
• Logistic regression with an L1 penalty
• Naive Bayes (GaussianNB)

The data was structured such that each row represented a flight with distance, departure + destination airports, total delay (in minutes), and weather information (this was merged from another dataset). I converted the delay column into a binary vector where 1 represented a delayed flight by 60 or more minutes, and 0 the flight was delayed by less than that threshold. This binary delay column became the y vector I was trying to predict and the rest of the data became the matrix of feature vectors, X.

After training the three models outlined above on this data, I found that the two logistic regression models performed identically at predicting delays of > 60 minutes. While Gaussian Naive Bayes model performed the worst. Looking at the feature weights of the logistic models, the L1 penalty logistic model used precipitation as the greatest indicator of delay, and the L2 penalty logistic model decided temperature best indicated a delay.

Packages used: Pandas, Scikit-learn, and Numpy.