About
Hi! My name is Hillary Chang. I am a performance engineer at NVIDIA, where I design and execute performance benchmarks for distributed AI workloads on GPU clusters, profile GPU utilization and interconnect latency, and build automation pipelines for large-scale benchmarking and analysis.
Prior to that, I built and operated core cloud infrastructure at Oracle Cloud Infrastructure, developing services and tooling for the Service Gateway, owning on-call across 200+ global regions, and leading deployments for distributed control plane services.
I hold a Master's in Data Science from UC Berkeley with a concentration in AI/ML, and a Bachelor's in Data Science with a Mathematics minor from UC San Diego. I am passionate about the intersection of high-performance computing, machine learning systems, and scalable infrastructure.
Experience
Projects
Full-stack ML system that classifies raw bank transactions using NLP, engineers behavioral features, and predicts credit risk. Achieved 98.93% transaction classification accuracy and 0.84 ROC-AUC on delinquency prediction.
Exposing the Limits of Transfer Learning
Research demonstrating that TF-IDF outperformed fine-tuned BERT by 8% on domain-specific tasks, exposing negative transfer from domain mismatch between Amazon reviews and banking intents.
Interactive educational website using D3.js visualizations to make Olympic sailing accessible to newcomers, featuring a global choropleth of Olympic participation, interactive boat diagrams, and a racing rules matching game.
Multi-Model ML Pipeline for Airfare Estimation
Machine learning system predicting flight ticket prices using multi-airline data, comparing Linear Regression, Random Forest, and XGBoost to achieve a 43% error reduction over baseline.
Competition Winner, Top 5 of 300+
High-performance Spanish vs. French language classifier ranking top 5 out of 300+ competitors, built with 240+ engineered linguistic features on an 800,000+ row dataset.
ML pipeline predicting calorie content from recipe attributes, achieving 99.54% R-squared accuracy with Random Forest regression on 83,000+ recipes from Food.com.
Statistical analysis of 83,782 Food.com recipes investigating the relationship between protein levels and calorie counts using hypothesis testing and missingness analysis.
Full-stack travel logging platform with user authentication, RESTful APIs, and complete CRUD functionality built with a team of 4.
Archive
Recognition
Skills
Personal
Outside of engineering, I love staying active. I rowed NCAA Division I at UC San Diego, played competitive tennis (USTA NorCal #3, 4-year varsity top singles, 2-year captain, 3-year MVP, nationals competitor), swam varsity, and played basketball competitively. I also play the Chinese bamboo flute (dizi), certified Level 8 of 9 on the Central Conservatory of Music exam with Honors.
I love training for races. I completed a Spartan Race and am currently training for a triathlon and half marathon. Other hobbies include surfing, rock climbing, hiking, golf, reading, and writing.
I also love volunteering. I tutored special needs students in robotics and programming at Friends of Children with Special Needs, coached Special Olympics basketball, and worked as a certified lifeguard for over 3 years.
