SEEK: Stacked Ensemble of Expert Knowledge

SEEK is a machine learning/natural language processing pipeline that utilizes model distillation to try to increase the performance on domain-specific knowledge of a large language model. In laymans terms, SEEK is an attempt to make large language models smaller and smarter.

Split.it

Split.it is platform that allows JHU students who use ride-sharing apps to easily find people to split the fare with. The app will leverage the social trust present in the Hopkins community to make organizing cheap transportation simple. I am responsible for all of the backend architecture, as well as integration with external services. I also built the backend for the fully-integrated chat. I also wrote most of the documentation for the project. Check out our project here!

Image Inpainting with Machine Learning

For this machine learning project, my team of four built and trained a Generative Adversarial Network (GAN) for the task of image inpainting, also known as image infilling. As you can see from the images, we mask the image with a square in the center, and the model generates the images on the right. We compared our model to analytical baselines, namely Navier-Stokes and Fast-Marching. The metric we used to quantify performance was Fréchet Inception Distance (FID). Our model performs quite well, and you can try it out yourself here.

Quest2Learn

Quest2Learn is an augmented reality (AR) application that enhances both online and on-campus lab sciences education. Our team has created an app in development with exciting technology using augmented reality and game design to teach courses in lab sciences. Currently, we are developing lab units with Johns Hopkins University (JHU) faculty members. Through Quest2Learn, we aim to address the lack of access to hands-on lab learning during the pandemic. My role in Quest2Learn was to develop these labs in the Unity (C#) environment. Find out more in our Notion page here.

For older projects, click here.