Engineered a production-ready AI orchestration pipeline leveraging LLMs and RAG context to automate system incident triage in a distributed environment. Collaborated within a 17-member software engineering team utilizing automated test harnesses and GitHub Actions CI/CD to validate system stability.

Engineered a high-performance image processing library that uses shared memory tiling to minimize global memory bottlenecks and maximize GPU throughput for convolutional computations.

Engineered a state-of-the-art anomaly detection pipeline in PyTorch, by writing the Patch Description Network class, training the teacher network and removing bottlenecks from the student/autoencoder training algorithm to allow for 70,000 training iterations in under 2 hours. Achieved an AUROC score of 0.88 on a diverse industrial dataset with less than 1 millisecond performance.

Engineered a regularized logistic regression model for breast cancer diagnosis, achieving a 98.59% accuracy on the WDBC dataset by implementing a regularized softmax cost function and gradient descent from scratch.

Trained a Random Forest model to classify whether or not an event is likely to be an exoplanet using NASA's Kepler data for the NASA Space Apps Hackathon.

Developed a ray tracer capable of rendering 3D animations using core vector math and rendering logic without relying on graphics APIs by following a textbook.

Engineered the hardware architecture for an STM32F4-based traffic simulation, integrating a custom circuit with 22 LEDs, a potentiometer, and 3 daisy-chained shift registers. Synchronized the 19-bit LED array via shared GPIO pins and ensured circuit safety by calculating and wiring precise series resistors to protect the power supply and components.

Engineered an emergency department, virtual triaging web application that allows patients to wait for an available physician from home.

Engineered an IR light detection robot from scratch, capable of detecting an IR beacon, travelling towards it and depositing a payload.