AI Engineer

AI engineering is not a single skill. It spans model research and training, infrastructure for serving models at scale, data pipelines that feed those models, and the application layer that makes the output useful. Most engineers are strong in one or two of these areas. I have built systems that touch all of them - by necessity, because shipping real products requires the whole stack to work.

My background is deliberately broad. I have done ML research resulting in peer-reviewed publications, built production pipelines processing millions of records, designed inference infrastructure routing across eight LLM providers, and shipped four complete AI-native products from zero to deployed. That breadth is not unfocused - it is the result of caring whether things actually work, not just whether a model achieves a benchmark number.

The through-line is an engineering mindset applied to AI: clear interfaces, observable systems, reproducible experiments, and graceful failure modes. A model that performs well in evaluation but fails silently in production is not useful. An AI product that works once but cannot be maintained is not shipped.

I have four peer-reviewed publications in IEEE Access, Springer, and Inderscience journals, covering medical image analysis, signal processing for clinical classification, and computer vision applications. The research was not done in isolation - the same pipelines that produced publication results were the foundation for production medical AI systems.

At ASU School of Sustainability, I currently engineer geospatial computer vision pipelines that process multi-year NAIP satellite imagery to analyze photovoltaic infrastructure at county scale. The work requires production-grade engineering: spatial fidelity quality gates (PSNR, SSIM, RMSE), reproducible processing across large image archives, and outputs that meet research publication standards. This is the gap most researchers do not close - between a result that is publishable and a pipeline that can be re-run by anyone on new data.

At Omdena, I led ML engineering for a platform processing over one million records, improving model accuracy from 72% to 89% and cutting preprocessing latency 40%. Leading the 15-person team required translating research decisions into engineering requirements that the full team could execute against.