About me

I work across technical product development, engineering, and AI/ML research/alignment, and currently study computer science at Columbia University.

Before returning to school, I built my career at Amazon, joining without a college degree and becoming the company’s youngest Level 6 (L6) at 21. I progressed from frontline operations into regional leadership and senior product management. At Amazon Business, I led ML products that ranked B2B products, forecast demand, positioned inventory, and generated replenishment recommendations for sellers. In Amazon Creators, I led session-level purchase attribution, ML risk controls for referral abuse, and computer-vision assessment of creator videos for enforcement and personalized recommendations.

I returned to Columbia to deepen my technical experience in AI and machine learning engineering and research. Through coursework and research, I am developing the foundations to implement models, design controlled experiments, and evaluate their behavior.

Most recently, I worked as a Technical Product Engineer Intern on the Starlink Enterprise Product Architecture team at SpaceX. I led product strategy and customer discovery for AI-cluster networking, helped define pricing and service tiers for dedicated internet access product and led engineering delivery of a unified Starlink business ordering system across service plans and hardware, including catalog redesign and satellite-capacity modeling.

I am completing a B.A. in Computer Science and a minor in Political Science at Columbia, with expected graduation in May 2027. My technical work includes controlled comparisons of CNNs and Vision Transformers, collaborative research on PCA and linear autoencoders, and prototypes for language processing, evidence-linked captions, and text distribution analysis. I am also developing a high-performance machine learning project on scalable LLM serving and inference optimization.

My research interests include efficient and scalable machine learning, multimodal representation learning, and reliable reasoning and agent systems. I am also interested in designing and building AI products for human use: translating user tasks into model requirements, designing interfaces and workflows, and evaluating whether the product helps people complete those tasks. I want to understand how model architecture, training objectives, and inference methods affect accuracy, memory use, latency, and performance on unfamiliar inputs. My alignment interests include mechanistic interpretability to trace the internal computations behind model outputs, and scalable oversight to test how people and automated checks can detect errors in tasks that are difficult to verify. I want to evaluate whether models remain accurate, acknowledge uncertainty, and follow user constraints when they encounter new tasks, unfamiliar language, or user needs that were not represented in their training data.