data science @ northeastern open to work
A benchmark of 300+ origami folds built to test how well vision-language models reason about space — whether a model can tell what a sheet of paper becomes after a given fold.
An AlphaZero-style connect 4 engine that learned entirely from self-play — a policy-value network guiding Monte Carlo tree search. The trained weights are exported to the browser, where a JavaScript reimplementation of the network and search plays every move.
An ensemble that detects AI-generated text by combining three models that read the same passage differently — a fine-tuned transformer, a stylometric classifier, and a CNN that renders text as an image. Quantized to int8, the ensemble runs entirely client-side — the text you paste never leaves your browser.
A pipeline that learns two decades of NBA MVP ballots to predict the winner of the 2024–25 season. it models each player's share of the vote rather than the result, so it's scored on how it ranks the field.
A tool that makes unfamiliar codebases readable by indexing them twice — a vector store for semantic search, a graph for structure — then answering questions with citations to real file paths and line numbers.