Hillary Chang

Design and execute performance benchmarks for distributed AI workloads (LLM training and inference) on multi-node GPU clusters using CUDA, NCCL, and PyTorch

Build Python and Bash automation pipelines to orchestrate large-scale benchmarking and performance analysis across GPU architectures

Analyze RDMA/RoCE networking performance to optimize multi-node GPU communication and scaling efficiency

Perform low-level GPU profiling with Nsight Systems and Nsight Compute to identify bottlenecks in utilization and interconnect latency

Engineered core infrastructure for Oracle Cloud's Service Gateway, improving scalability and fault tolerance for high-throughput enterprise networking systems

Owned on-call for 200+ global regions, resolving Sev1/Sev2 production incidents using Grafana dashboards, Lumberjack logs, and internal debugging tools

Led global deployments and region build workflows, managing rollout, validation, and rollback for control plane services

Debugged complex distributed system failures including Terraform state drift, API throttling, and network misconfigurations across multi-region environments

Developed 30+ cross-platform CMake modules for STEP Adapter and Parasolid-STEP Translator, enabling CAD file translation across Windows, Linux, and Intel/ARM macOS

Unblocked Linux and macOS development for the CAD translation team by delivering the first cross-platform build system, replacing a Windows-only proprietary toolchain

Diagnosed and resolved 10+ platform-specific build failures, including missing Perl-generated headers, macOS compiler incompatibilities, and linker symbol errors, in a legacy C++ codebase with no existing CMake precedent

Traced and translated 20+ library dependency graphs from proprietary .def files into CMake targets, resolving linker, compiler, and header generation issues across three OS toolchains

Built a code style checking tool enforcing DO-178C aviation safety standards, adopted by 1,000+ developers across engineering teams

Automated 100+ critical validation tests for VxWorks real-time operating system, reducing manual QA efforts and error rates

Integrated regex-based code parsing and Doxygen/ApiGen documentation standards to automate code quality validation across the codebase

Engineered a real-time data streaming pipeline using AWS Kinesis, improving data throughput by 20% for indoor localization research

Configured and deployed 10+ Raspberry Pi units for continuous data collection and secure real-time tracking of localization devices

Defined advanced analytics and ML project objectives in collaboration with faculty, translating sensor data into actionable insights for wireless sensing research

Engineered a YOLOv5 deep learning model achieving 90%+ accuracy on image recognition by training on COCO and custom datasets

Built a web crawler using YouTube API and FFmpeg to download and preprocess video data, enabling large-scale dataset collection and analysis

Automated extraction and preprocessing of 7,000+ images with data augmentation techniques to improve model robustness and generalization

Fine-tuned hyperparameters using precision, recall, and F1-score metrics to optimize detection accuracy across diverse datasets

Applied Kalman filter signal processing to denoise 30,000+ rows of sensor data from Fiber Bragg Grating (FBG) optical systems

Proposed and integrated multi-sensor fusion methods to improve accuracy across multiple FBG sensor channels

Developed MATLAB visualization scripts to analyze sensor data patterns, creating plots and graphs for engineering review

Presented technical findings and strategic product recommendations to the CEO and senior management, informing future R&D direction