At U.S. Bank, I created a custom pricer for commodity swaptions with flexible correlation matrices, enabling accurate pricing and trading of the product. I also updated a futures options pricer to handle volatility and debugged the Aegis-to-Calypso trade-loading pipeline to remove pricing errors.
At Linksome, I developed text classifiers, with a logistic regression model that achieved 95% accuracy while matching LLM performance at lower cost. I also built asynchronous crawlers that doubled download throughput and a text-processing pipeline that extracted 15% more relevant data.
At GAINS, I built a travel-time prediction model with a spatial data augmentation pipeline, reducing API costs through proximity-based greedy polling. In my LLM + XGBoost House Price Prediction project, I developed an end-to-end system combining structured and visual features, including a custom Zillow crawler and natural-language feature extraction.
