Highly motivated to use technology to create real impact in the communities around me. Always excited to grow, learn, and expand my skills in new directions. I have worked with Python, Golang, C++, machine learning, and AWS services.
A paper released a few years ago from MIT proposed an alternative solution to sleep tracking: determining sleep stages with a ML model using radio signals. We implemented this model, and achieved a fairly high accuracy. I experimented with implementing a ResNet model to be our encoder. I also helped create a website to display the data using React.
Fall 2021 Project: Scribe
We constructed a model which has the ability to generate images of a user’s handwriting using very few samples of the handwriting by combining the ScrabbleGANN and Reptile architectures. We added an extra discriminator to ScrabbleGANN called a “Recognizer” which pushes the GANN to generate text in the same style as a specific user.
Spring 2022 Project: Launchpad x DroppTV
I worked with a startup, DroppTV, to create a model to effectively classify a shoe’s brand and model from video data. I experimented with self-supervised contrastive learning, which aims to create an embedding space where similar samples are close together. I also tuned training hyper parameters to increase model performance and generalizability on two separate datasets.
I improved an entity classification model for wit.ai, a platform to build natural language understanding models and re- solved long-standing user errors. To do this, I implemented the new model architecture, along with its loss, metrics, and torchscript conversion methods. Next, I designed and created a training pipeline which fetches data, trains the model, and publishes the model to production. Finally, I integrated the model into the wit.ai backend and deployed the model to Czech-speaking users.
We are working to improve autonomous control of floating seaweed farms using ocean currents.
I created a lightweight and accurate model of seaweed growth to aid in planning for optimal seaweed growth, and I developed the model using the jax.numpy library and integrating with the simulation hosted on c3.ai’s platform. Now, I am exploring different methods of planning a route far in the future without accurate temperature forecasts.
As a reader for CS 170, I helped students with their homework in office hours. I also helped create rubrics and graded the students' homework and exams.
