Data-Driven Modeling and Studies

System modeling leverages experience and intuition to identify structure. Modeling exposes hidden relationships, guides trade studies, and enables decisions that would otherwise rely on guesswork. My work spans decades of modeling—from physical systems and sensor geometries to modern unsupervised learning and civic data analysis.

I modeled the structure of the rental property market in Cambridge, MD using public records, GIS overlays, and licensing data. That work shaped the framing and recommendations of a code enforcement and rental licensing study. The study reflects policy and organizational analysis, from insights that emerged from structured data modeling—notebooks built in Python and Google Colab that revealed how a small number of owners controlled large shares of the market, and how licensing gaps aligned with compliance and inspection patterns.

Other projects featured here reflect a broader evolution in my modeling practice.

I modeled radio propagation and terrain-constrained VHF communications to infer interaction networks—deriving unsupervised insights about coordination, isolation, and peer group structure. That project introduced me to graph-based learning, and showed how ML methods could extend physical intuition.

I’ve also applied clustering and dimensionality reduction techniques to classify civic and operational data where labels were unavailable or insufficient. These models support triage and inference in real-world workflows—identifying structure where others see only noise.

Each case, reflects a modeler’s mindset—one that leverages my background in estimation theory, sensor modeling, and workflow design.

These studies support policy, triage, automation, and organizational insight. They enable workflows where agents can classify, group, notify, or escalate based on modeled structure—not rigid rules.

Agentic AI depends on modeling. It starts with understanding the system, the data, and the constraints. These projects reflect that foundation.

  • This study integrated Maryland real property (SDAT) records, licensing databases, and GIS overlays to analyze housing code enforcement and unlicensed rental risk.

    This work explored methods like k-means, DBSCAN, and PCA to find latent structure in mixed civic, behavioral, and observational data. The work was done using Python 3 in Google Colab, with visual outputs in Google Data Studio.

  • My foundation in structured estimation came from work in radio direction finding, using array manifolds and geometric projection techniques to estimate signal direction. These methods required modeling high-dimensional spaces and extracting structure without direct labels—mirroring unsupervised learning in principle, long before modern ML frameworks emerged.

    This work shaped how I think about latent space geometry and explainability in agentic AI.

  • I applied unsupervised learning to VHF radio C4ISR collection data across complex terrain. Using terrain-informed propagation models, translator gists, and modularity clustering techniques, I derived a social interaction graph of radio users—inferring community structure and regional collaboration patterns.
    Employing a combination of domain-specific physics and ML-driven pattern discovery provided valuable experience in graph-based analysis and community detection.

  • People don’t engage with technical detail or policy unless visuals connect—and the narrative is intuitive. I have deep experience blending workflow concepts and model awareness in clear, audience-targeted explanations. Good models are only useful if others can understand and act on them.

    This example presentation breaks down complex interactions—like inspection timing, license triggers, and remediation escalation—using humor, visual framing, and step-by-step illustrations.

    Annotated tables, bulleted comparisons, and data-backed charts highlight patterns. Flow diagrams show how the licensing process feeds structured data back into the city’s policy feedback loop.