A selection of work spanning shipping Apple features, large-scale consumer and AR apps, and graduate machine-learning research.
Multi-process, multi-threaded sensor-fusion pipeline letting an Apple Watch locate an iPhone via triangulation. Implemented moving/stationary and multi-floor detection, plus internal tools to log and replay finding sessions — cutting phone-search time by roughly a third.
Designed and implemented a C++/assembly firmware framework for static and motion classification on custom Apple silicon, consumed by the Audio and AirPods teams. The framework provides device-state decisions at varying sample rates, enabling the Audio team to light the AirPods charging LED the correct color when the device is placed in its case.
Designed and implemented the pocket-detection algorithm and SPI for iPhone 16 / 16 Pro's Camera Control, deriving a low-power solution from Always On Display analytics to avoid accidental activations.
Designed and implemented the pocket-detection algorithm for iPhone's Always On Display feature — analyzing luminance and sensor-fusion data to suppress the display when the phone is pocketed, saving battery without sacrificing usability.
Deep learning model for multi-class cancer classification from histopathology images, combining convolutional feature extraction with ensemble techniques to improve diagnostic accuracy across multiple cancer types.
Architected and built AR frameworks for Royal Caribbean cruises: marker-based games recreating classic fair and strategy games (Two70), and an X-ray experience streaming 360° captain's-view video when pointing an iPhone at markers on the ship's walls.
Developed and maintained Disney World / Disneyland app features: an Instagram-style PhotoPass gallery with photo editing, the shared Disney Core UI and services layer, and Memory Maker payments, transactions, and park ticketing.
Built a Swift iPad scheduling app for 20,000 United Airlines pilots: a full-screen dashboard of current and upcoming flights, flight plans, vacations, airport door codes, delay notifications, bulletins, and crew information.
Graduate deep-learning project: a VAE-LSTM model in PyTorch predicting vehicle motion, inspired by Social LSTM and Social GAN. Built a custom multi-agent dataset from Lyft Level 5 data feeding multiple agents' positions per frame.
