At Columbia University, I design and conduct empirical studies of how mathematical concepts are represented in Large Language Model activations to corroborate the Linear Representation Theorem. I also built a large mathematical language corpus for representation-learning experimentation.
At Meadow AI, I improved the precision of a computer vision model used by 21 customers by 30%. I also optimized a YOLO model for a difficult detection task with limited data and computing resources, increasing mAP50 by 42%.
At Shield AI, I trained reinforcement learning agents to cooperate and delegate responsibilities using PyTorch. I also initiated and built a repository of shared testing utilities in response to a corporate need.
At the University of Illinois Urbana-Champaign, I planned and executed experiments on natural language processing tasks including language detection and text summarization. I published findings in IEEE/ACM Transactions on Audio, Speech, and Language Processing and presented at the International Conference on Intelligent Computer Mathematics.
