Projects

• Compared classification improvements of various data augmentation techniques on a Spotify lyrics dataset and a Linkedin Job Postings dataset. Both included significant class imbalances.
• Implemented random swapping, insertion, deletion and synonym replacement and used them to augment both datasets. Other augmentation techniques were back translation, and LLM prompted augmentation.
• Fine tuned a pre-trained BERT model and implemented and trained a BOW SVM in Pytorch on all augmented/original sets, compared f1 score performances, and generated t-SNE plots for visualization.

• Implemented double DQN and PPO from scratch using Pytorch
• Trained agent using DDQN on cartpole and lunar lander, converging consistently at max return for cartpole and frequently over 200 on lunar lander.
• Trained agent using PPO on bipedal walker and lunar lander, converging at near 300 return on bipedal walker and near 200 on lunar lander.

• Worked in a team of 9 to design and create the web app Punch Time, an hourly employee time-tracking tool.
• Designed parts of the data model and API and set up endpoints to support CRUD operations using Express and Mongoose to communicate with a MongoDB cluster.

• Worked in a team of 4 to design and create the web app Invego, a stock market-like simulator where users manage and sell their personal stock, as well as buy and trade the stocks of other users.
• Contributed to Figma designs and HTML implementation, utilizing CSS styling and animations. Wired up routing between pages, integrated front and backend using Javascript fetch API, and updated DOM accordingly via Javascript.