- I am currently an undergraduate student majoring in Computer Science & Mathametics
- My technical experience includes:
- Software Engineering Intern @ Truss
- Software Developer Intern @ University of Michigan
- Deep Learning Research Assistant @ Eastern Michigan University
- IT Intern @ Compass Group
- My main areas of expertise are:
- Building web applications using Ruby on Rails
- Machine Learning, specifically Transformer models in Natural Language Processing
- AWS for cloud services for system design & practical ML
A Rails application built for the LSA Sight and Sound team to streamline the daily inspection process for 300+ classrooms accross University of Michigan.
A tool built at GrizzHacks 6 with a team of 4 that won the AI/Machine Learning track and the award for the best .Tech domain name. GrizzScribes converts lecture material (video lectures, audio recordings, powerpoints) into (1) markdown notes and (2) flashcards.
An alternative to EMU's course search that allows you to look up courses based on availability schedules. It consists of a full-stack application hosted on AWS, with a PostgreSQL database, a Ruby on Rails backend (api) and a JavaScript frontend.
Research project that utilizes outside context to improve the performance of source-code Transformer models. Using PyTorch and the Hugging Face Library, architected and trained models on the CodeSearchNet dataset.
Cohort project built for the Eisenberg Family Depression Center at University of Michigan Medicine. Enhances the current CRM-style application with Sharepoint enhancements, PowerAutomate solutions, and Tableau dashboards.
A financial micro-learning application built at MHacks 16 with a team of 4. It includes an article recommendation system that uses item-based collaborative filtering, and an article retrieval and processing pipeline that integrates GCP, MongoDB, and GPT.
A transformer model trained on German eBay listing titles for Named Entity Recognition. Model pretrained on an MLM task on 10,000,000 examples, and further fine-tuned on a downstream NER task. After conducting hyperparameter search, optimal model achieved 86% F1 score in the final competition ranking.