About

PurcellAI

• ▸Implemented a full-stack data pipeline: from low-level serial capture on embedded devices to higher-level processing and orchestration in Python.
• ▸Developed firmware for Arduino Nicla Vision (Cortex-M7 MCU + onboard sensors), focusing on low-level embedded programming in C++ and Mbed OS.
• ▸Deployed quantized models on Nicla Vision, achieving 90% accuracy with real-time (120 ms) on-device inference for guided detection tasks.
• ▸Implemented manual SPI communication routines to interface with onboard sensors, exploring bit-level timing, protocol handshakes, and register access.

LongShot Space · Oakland, CA · Contract

• ▸Designed custom parallel data structures and optimized algorithms to support CUDA-accelerated 1D explicit simulation of a booster-aided light gas gun, achieving a 50× runtime reduction over CPU execution.
• ▸Developed and optimized CUDA kernels for parallel execution, building scalable pipelines capable of running 500+ concurrent simulations on a single GPU and extending to multi-GPU deployments.
• ▸Built GPU-accelerated ML pipelines with RAPIDS and TensorFlow, processing 100,000+ simulation outputs to streamline training, inference, and optimization workflows.
• ▸Benchmarked and profiled CUDA kernels, improving memory throughput by ~30% under large-batch, distributed simulation workloads.

DragonFly Energy · Reno, NV

• ▸Used spectroscopic tools (FTIR, RAMAN, QCL) to design new methods for battery degradation analysis.
• ▸Researched and determined battery failure mechanisms including delamination, dendrite formation, gaseous byproducts, and poor SEI formation.