×
- Built a full-stack cross-platform mobile app to help users track, maintain, and leverage their network of personal contacts.
- Developed frontend using react-native on the expo framework, using Tamagui components styled with centrally defined design tokens as well as with Nativewind/Tailwind.
- Developed backend as a python flask app running on AWS Elastic Beanstalk with a Postgres database hosted on Neon.
- Implemented natural language querying for contacts using semantic search over OpenAI embeddings of contact profiles with pgvector, and implemented geographic proximity sorting using PostGIS.
×
- created market sentiment prediction program that scrapes recent tweets to classify any stock as bullish, bearish, or neutral with approximately 50% accuracy.
- Implemented using custom and out-of-box prediction models from Tensorflow and scikit-learn.
×
- Worked with interdisciplinary team of Masters and PhD students to create visualizations to interpret and understand text classification models.
- Distilled complex NLP models into 1st and 2nd order linear approximations and compared these to transformer attention mappings.
- Evaluated each approximation and visualization method and summarized results in a symposium poster.
×
- Led NLP project team working on using RNNs and transformers to generate text in the styles of particular authors.
- Coordinated and contributed to development of mutliple sucessful RNN and transformer-based models.
×
- Creating chatbot that uses seq2seq architecture to intelligently respond to user prompts.
- Implemented using custom Tensorflow LSTM models.
×
- Initiated, planned, and executed a project to compare common RNN-based models on their ability to generate Reddit comments in styles of three subreddits in response to a prompt.
- Compiled and synthesized almost 100 academic papers into executive summaries in the meta-research phase.
- Modified code from an ACL-2019 paper on LSTM-based seq2seq comment generation with user feature embedding using TensorFlow.
×
- C++ library of two-sided versions of common sorts with templated support for unspecified containers and datatypes.
