Technical work spanning AI, machine learning, and software engineering
Developed an automatic system utilizing LLMs to semantically parse media archives from RSS feeds and interactively engage users with reliable, real source attribution. Primarily worked on retrieval search mechanisms (RAGs), LLMs, database management, and product strategy. Oversaw multiple live tests in partnership with NPR.
Multi-agent reasoning chain-of-thought framework to improve code policy generation for robotic tasks, building upon previous Code as Policies research.
A MDP modeling approach addressing the sequential decision-making problem of recommending books to users based on their historical interactions.
A comprehensive study on the development and evaluation of an LLM-driven tool designed to assist in the grading and constructive feedback of Common App personal essays.
Investigating the potential of Low Rank Adaptation (LoRA) for style transfer text generation, with the aim of emulating the writing styles of prominent politicians.
A novel methodology combining pose estimation techniques with deep learning models to accurately classify fitness exercises.
Building on OpenAI's powerful generative language model to optimize a RAG for Planet Money's podcast archive.