Computer Science at the University of Illinois Urbana-Champaign. I write battery-management firmware for an electric race car, build iOS and macOS apps, and do machine learning research on speech recognition.
Write real-time C firmware on FreeRTOS for the race car's battery management system (STM32H7): a 1 Hz CAN heartbeat and the charging algorithm for the 140-cell pack, which tapers current as the highest cell nears its voltage limit. Build with Bazel, flash through a CAN bootloader, and verify on the bench with a CAN bus analyzer.
Led development of Model mAKe Smith, a platform that ports, runs, and compares AI models on remote edge hardware through chip-vendor toolchains (starting with Cadence NeuroWeave), with side-by-side speed and accuracy results. Also led development of Model Analyzer, a free in-browser tool that inspects ONNX, TFLite, Keras, PyTorch, and TensorFlow models layer by layer, with weights and multiply-accumulate cost, without the model ever leaving the browser.
Diagnosed and resolved defects before deployment and optimized existing code paths; cleaned and analyzed large datasets and tightened collection methods to reduce analysis error.
Developed AI software for edge computing on microprocessors and built an AI-powered subject-tracking camera that kept tracking beyond the frame boundaries.
Built a voice-activated GitHub FAQ on Alan AI's speech recognition platform, designed to cut support wait times by about 30%.
Sole developer of the full-stack platform a daycare runs on every day: an iPhone app for owners and teachers, iPhone and Android apps for families, and a web portal for billing. Daily reports, photo sharing, attendance and pickups, messaging, PDF invoicing, e-signatures, and an AI assistant for development plans.
lotusbudacademy.com/software →iPhone and iPad app that turns notes, scanned pages, PDFs, slides, and Google Drive files into flashcard decks, written on-device with Apple Intelligence.
View on GitHub →Fully on-device face recognition for macOS: Vision detection, 5-point alignment, Core ML embeddings with cosine matching, and blink-based liveness to reject photos.
View on GitHub →macOS battery menu bar app that reads live current and voltage from the SMC, iPhone and iPad battery over Bluetooth LE, and AirPods levels from Continuity broadcasts.
View on GitHub →A Dynamic Island for the MacBook notch: Now Playing, timers, AirPods and device batteries, calendar countdowns, and file drops.
View on GitHub →Evolutionary ML experiment: 20 small neural networks learn online from live Coinbase order-book data, evolve by mutation and tournament selection, and are judged against coin-flip twins. Paper trading only.
View on GitHub →Co-first author of Evaluating the impact of playback speed for transcription accuracy, under review at the Journal of Emerging Investigators. Benchmarked four speech recognition systems (Whisper, Azure, Deepgram, Google Cloud) across 150 hours of audio and four playback speeds, and found model-specific speed thresholds that cut transcription cost about 20% without losing accuracy. Also built an iOS app and an AI Google Chat bot used by 30+ lab groups and 1,100+ student researchers.
A wearable foot sensor that uses machine learning on gait data to detect early signs of osteoporosis, built around a Raspberry Pi and a 16-sensor array with a real-time analysis pipeline.
Discrete Structures, Linear Algebra, University Physics: Mechanics
Structure and Interpretation of Computer Programs (CS 61A), Data Structures (CS 61BL), Foundations of Data Science (Data C8)
Founded a 501(c)(3) nonprofit that teaches computer science to 50+ underserved students; built the curriculum and manage volunteer instructors.
Directed operations and coached 100+ students in a program serving 7,500.
Senior Patrol Leader and National Youth Leadership Training staff; Presidential Volunteer Service Award (Gold), twice.