Hi there 👋
I lead AI infrastructure solutions engineering at Cisco, running a team that covers enterprise AI and GPU cloud accounts across the US West. I primarily focus on large buildouts: cluster and fabric design, capacity planning, and the business case to link the infrastructure to the funding. A lot of my time goes to the economics, because that's usually where the decision eventually ends up.
I study technical specifics and read analyst reports on fabric architecture, serving stacks, the hardware and software boundary. Most of my job is spent designing buildouts or identifying benefits and drawbacks of AI infrastructure decisions.
Before Cisco I finished a PhD in physics-informed machine learning and spent four years in federal AI. Most of the code here is from that period.
SA-PINNs - Self-adaptive physics-informed neural networks. This was the bulk of my dissertation work. Trainable adaptation weights applied per collocation point, so the network learns which regions of the solution are hard and concentrates there. The paper is at 1,250+ citations and the method has been picked up in several downstream PINN libraries.
TensorDiffEq - Open source framework for multi-GPU training of PINNs over large problem domains. Built for scientific ML workloads that don't fit on one GPU, with materials applications as the primary target.
Multimodal transfer-learned regression - Computer vision applied to materials property prediction.
I contributed TensorFlow implementations and cross-framework technical content to d2l.ai, the open source deep learning textbook now adopted at 500 universities in 70 countries and published by Cambridge University Press.
Rated Army Reserve fixed-wing aviator. I've reviewed technical manuscripts for Manning, mostly on distributed computing and TensorFlow.