Generates custom Japanese study decks by filtering vocab lists on frequency, minimal pairs, and learner status. Built to give pitch-accent learners structured practice based on guidance in the NHK Pitch Accent Dictionary.

• Ships with accent data compiled from various sources, and a Yomitan dictionary to match.
• In the best case, can make learning pitch accent 30% faster.
• Allows users to focus on language practice that matters more.

My team's senior capstone project, focusing on combining iNaturalist observation data with GIS data from various sources to provide insights on the distribution and habitats of hundreds of thousands of species.

• Populate description/field guide for species automatically.
• Provide a list of flowers that are in bloom nearby, or birds that are migrating through the region, for example.
• Planning on migrating to an open-source project upon the conclusion of our senior capstone.

A project focusing, on a high level, to address the pervasive problem of annoying, unhelpful, and even downright harmful ads encountered while watching live TV, such as a sports broadcast. The recent surge in unregulated sports betting advertising was an impetus in the creation of this project - yes, I'm looking at you, Kevin Hart and LeBron. See YoutubeOverCommercials for a solution that works quite well. I want this to work even if there is no reliable logo to detect on the broadcast, and also to explore how to fight pervasive advertisements using AI in general.

• Take live transcription and classify sentences into broadcast content or advertisements.
• Somewhere down the line, incorporate image processing to detect visual advertisements and ad banner boundaires.
• Train a classifier to take this kind of data as input, and hide ads with an overlay and/or muting.
• Optimize the model enough to the point where it doesn't take a 4000 series NVIDIA GPU to run it.