Developing a model to detect defects in helmets used by pilots. Also creating a frontend environment for the model to run on and display the results.
Performed exploratory data analysis on data from a manganese production facility. Trained a Variational Autoencoder to reduce a 16,538 dimensional vector to 2 dimensions. Analyzed 2-d vector to determine clusters that maximized production output. Integrated data preprocessing and data report pipelines with the frontend using Django and React.
Created various models to diagnose Parkinsons patients from datasets, using xgboost and InceptionV3 transfer learning. Presented results at UMD's undergraduate day and the MLSE Conference at CMU.
A dynamic website that allows users after they signup to add campgrounds near their location, reviews of the campgrounds, and general comments. Using a non-relational database users are able to connect with their campgrounds and comments that are saved in MongoDB. In addition, the website uses RESTful routes to create, update, destroy, and show campgrounds and comments.
A fun static website that allows a user to play sounds from the keyboard on the website. Paper.js is a javascript library that provides the animation for the circles when a key is pressed, and Howler.js provides the functionality to play certain sounds. This idea came from https://patatap.com/ another website that allows a user to play default sounds on key presses.
A dynamic website that allows any visitor on the website to create, edit, or delete a blog. The blogs are stored in a non-relational database using MongoDB. This site uses Semantic UI which essentially is Bootstrap, but has more meaningful names for css classes. The idea behind this site was to practice RESTful routes, and attempt to try a new framework over Bootstrap.
A UC irvine dataset extracted features from the audio data recorded from various patients with parkinsons and without parkinsons. Using Scikit Learn I attempted to find the best accuracy given the dataset using various supervised learning techniques and got a 98% accuracy w/ KNN, and 96% accuracy with XGBoost. I also plotted the importance of the features in the dataset, and dropped features that had no impact on the tree model.
TwoSigma's Artificial Intelligence Challenge that tests a user's bot against other's players bot. Initially, began with a python implementation of strategies, and was placed at about 360. However, this began to exceed the amount of time I had for each turn, causing me to lose. Therefore, I switched to a ML strategy, using the ML starter bot I scraped data from the top players and trained the bot on the winners of those matches, reaching a rank of 401. Then I switched over to Java and back to algorithmic implementations where I was able to place 148. My github link on the right provides a more detailed description.
GPA: 3.63
GPA: 3.92
Computer Science
Mathematics/Statistics