John Hodge
Research scientist and engineer working on ML for physical systems.
I build models and tools for hardware reliability, RF and electromagnetic systems, engineering design optimization, and infrastructure-scale decision-making. The common thread is physical systems: problems where the data is messy, the failure modes are real, and the model has to survive contact with engineering judgment, cost, and operational constraints.
See the projects Resume Get in touch
What I work on
Applied ML for hardware reliability and sensing
Repair recommendation, failure diagnosis, and multichannel sensing turned into decisions on noisy operational data.
Physics-based engineering and simulation
Electromagnetic solvers, phased arrays, and antenna design, modeled in code and driven by an LLM agent.
Open-source tools for engineering decisions
Trade-study frameworks and optimization packages for system-level engineering decisions.
Featured projects
All projects →Hardware diagnostics and repair recommendations
Industry workProduction ML decision systems that fuse server telemetry, logs, and repair history into calibrated, component-level repair recommendations for AWS EC2.
- Python
- PyTorch
- SageMaker
Agentic Phased Array Builder
An LLM agent that orchestrates electromagnetic solvers and array models through the Model Context Protocol, running auditable phased-array workflows from unit cell to system metrics.
- Python
- LLM agents
- MCP
Phased array system trade studies
A requirements-driven Python framework for phased-array system design: DOE and Pareto trade studies over link budgets, radar detection, RF cascade, reliability, and cost.
- Python
- NumPy
- SciPy
My PhD work was on reconfigurable metasurfaces and generative deep learning for electromagnetic design. See the research, publications, and talks →
Beyond the work
Coffee, investing, reading, hiking and the outdoors, travel, and Duke basketball. I also build BeanBench, a specialty-coffee logging app for iOS.