• Developed a multi-class product classification model using BERT and Logistic Regression to classify product relevance as Exact, Substitute, Complement, or Irrelevant, achieving an F1-score improvement from 50% to 63%.• Processed over 100,000 product queries using NLTK and regex for stemming, lemmatization, and stopword removal, optimizing data preprocessing for multilingual NLP tasks.• Implemented data merging and preprocessing techniques using Pandas and BeautifulSoup to clean product metadata and remove improving model performance by addressing RAM overflow issues.

• Engineered a robust CNN model using Pytorch that achieved a 75.4% accuracy on CIFAR-10 adversarial dataset, achieving a 64.1% improvement from the baseline model• Led research project emulating human reasoning in 3D space with point clouds and transformer neural networks• Implemented salient mapping, gradient clamping, and dropping in Python for point cloud classification, raised accuracy by 30%

• Spearheaded automated report generator to streamline lab documentation; leveraged Google Speech-to-Text, OpenAI API, and Python to transcribe experiments and edit into a comprehensive lab report, optimizing writing processes in research labs by 90%• Redesigned two-folder file system into clear, hierarchical system with modular components, boosting team-wide productivity• Developed client portal using React and TypeScript to display of orders. Seamlessly integrated Mailchimp for automated email form submissions, pulling data directly from a Supabase backend• Integrated chatbot for lightning fast real-time customer support, enhancing user experience and support efficiency

Selected as 18/250 applicants for a year-long fellowship with capstoneDesigned an apartment design app. Coordinated a team of 3 to mock entire process of creating a new product, including conducting market research and user interviews, wireframing, and prototyping.

Directed a team of 5 consultants to work for SaaS client. Built Metabase queries with SQL to expedite their consumer analysis.Identified user trends using Python and Matplotlib. Recommended client top 6 ways to boost conversion across two deliverables. New update in the app reflected our recommendations.

Piloted a FullStory integration, an UI analysis SaaS product, on Woflow's platform. Created 5 client dashboards to pinpoint user friction occurrences for over 200 users.Investigated 12 self-service onboarding SaaS softwares, attending product demos with sales teams and tested various sample flows. Proposed top three best-fit softwares to the CEO to enable a better customer experience.Conducted competitive research on 16 POS and CRM companies to aid the Design team make informed design decisions.

• Co-authored Interpretable Personalization via Policy Learning with Linear Decision Boundaries (Prof. Zhengyuan Zhou)• Trained machine learning models (Random Forest, Nearest Neighbors, clustering, etc.) on 1 million samples of e-commerce data with Python, Pandas, and Scikit-learn to maximize sales. • Resulted in a 33.8% increase in net sales revenue; leveraged vectorization to optimize the models' efficiency by 10x

Worked with Professor Kevin D. Murphy @Vanderbilt to research the inverse relationship between popular Japonisme and the discrimination that Japanese immigrants have faced in the US. Worked with Lecturer in Sociology Dr. Ted Perlmutter @Columbia to study the change in the Slow Food movement’s public strategies from 2010-2018.

Managed data operations, led a team of five college interns on the development of a new app feature that indexed 840 new terms, held a promotional event at a local cafe, and managed seven freelancers. Also led long-form content creation and published Medium articles.