I am an AI Researcher and Applied Engineer (PhD Candidate, ABD) bridging high-stakes federal missions and advanced technical implementation. Currently Vice President of AI at US AI, I architect secure, human-aligned GenAI ecosystems and lead research into Security Normal Form (SNF) for canonicalizing large-scale security telemetry.

• Mission: To treat AI safety as an empirical science, using rigorous measurement to understand how models impact human wellbeing.
• Current Focus: Building Agentic Workflows, Graph RAG, and automated cATO patterns for federal GenAI adoption.
• Philosophy: "Understanding the 'latent' traits of ourselves and others is the key to breaking down barriers and fostering empathy."

My dissertation at the University of Tennessee, Knoxville (advisor: Jens Gregor) is APERTURE, a five-paper chain on measuring personality in language-model agents when no ground truth exists. The central finding: frontier models rating well-known subjects retrieve rather than measure, so a valid instrument needs supervision that cannot be retrieved. The record behind it is 129 pre-registrations with 31 refutations reported in full.

Forty-four psychometric models across seven traditions encoded as factor chains (6,694 chains, 358 factors) with embeddings, trained classifiers, a FAISS semantic index, and an expert-evaluation toolkit. 86.8% best-per-model accuracy on 418 human-authored items from 22 published instruments.

Six measurement methods on 562 literary characters across 75 works. A 25-rater panel spanning nine provider families reproduces the Honesty-Humility/Agreeableness fusion that HEXACO was built to separate, and that structure collapses on characters constructed to lie outside any training corpus. A cache-membership gauge separates retrieved from measured ratings at AUC 0.99. The repo holds the notebooks, data, and explainers to reproduce the paper in experiment order.

Ground truth designed into synthetic material so the label is true by construction and cannot be retrieved. The frozen, hash-pinned scene-routed HEXACO instrument recovers all six axes at r = 0.75 to 0.81 with test-retest 0.946. Turned around, it shows that a persona specified as conduct is recovered at r = 0.81 and holds for twenty turns, and that designed traits reach behavior on tasks with countable outcomes.

The same discipline applied to PHQ-9, PCL-5, CAPS-5, C-SSRS, AUDIT-C, and HiTOP: six instruments encoded, tier-0 calibration, a drafted clinician-validation protocol, and a tiered ethics framework. No patient-derived data. Expert input from a VA National Center for PTSD clinical psychologist.

Applied research at US AI to canonicalize heterogeneous security telemetry using GPU-accelerated density clustering on Kubernetes/Kubeflow: 98,639 findings reduced to 21,444 canonical clusters.

• Impact: Deployed the VA’s first GenAI production system on Microsoft Azure Government (MAG), scaling to 1.5M+ daily records for mental health detection.
• Innovation: Surfaces Social Determinants of Mental Health (SDoH) through closed-loop AI observation.

A complete graduate-level curriculum (COSC 650, UTK) covering the practical applications of Generative AI.

• Modules: RAG Foundations, Agentic Workflows, and Constitutional AI.
• Resources: 400+ curated research papers and 50+ hands-on notebooks.

• VA Innovation Award (2024): For the CLEVER Pipeline & AI-Assistant deployment.
• $1.34M AVIN Innovation Grant: Integrating personality models into autonomous systems.
• $1M ENCQOR 5G Grant: AI/ML behavioral integration in connected corridors.
• NSA Scientific Achievement Award: Critical mission research in boundary defense.

• LinkedIn
• X