In order of how interesting they are.
This project was presented at the NeurIPS 2021 conference, datasets and benchmark track. The aim of this living benchmark is to accelerate progress towards self-supervised AI models that can perform tasks well regardless of the type of data they are trained and tested on. My primary contributions were the implementation of the transfer layer used in our baseline models, and the project website and leaderboard.
This was a brief project I did at the tail end of high school (May 2020) where I used Google Scripts to automate the sending of personalized yearbook messages between students. In the end, around 350 students between 12 schools sent and recieved 2,000+ messages on their graduation day. The website is no longer up.
Before I knew anything about NLP, I made this website that basically counts the number of unigrams in common between two source texts and a target text to give the relative probability that the source text author wrote the target text. It's crude, but sometimes produces interesting results.