I find where AI belongs in government work,
and ship products people can trust.
Product and delivery leadership for AI in federal agencies. 12+ years building decision-support systems for NSF, USDA, USPS, and Census, plus a startup I founded. I do discovery inside the workflow, and I ship systems where the people accountable for outcomes can see, override, and explain every recommendation.
What I bring
Discovery inside the workflow
Panel Wizard started by watching program staff work across 8 screens and spreadsheets. The product consolidated them into 1, let sentence embeddings suggest review panels, and kept every decision with the staff. Panel formation went from weeks to hours.
AI product judgment
Ambiguous proposals go to people instead of being forced into clusters. Assistants have to cite their sources. Overrides are logged and measured, not hidden. I design for evaluation and human oversight from the first sprint, because that is what gets an AI product adopted in an agency.
Shipping through teams I do not control
Consulting means executing through agency staff, security and governance stakeholders, and vendors. I delivered a governed AWS data platform across Salesforce, CBP customs records, and investigative systems for 50,000+ USDA operations, then ran the study halls that made adoption stick.
Selected work
All case studies
Product build (Gettysburg tourism ecosystem)
Gettysburg Pulse: Real-time event discovery with trust, provenance, and deterministic AI workflows
Built Gettysburg Pulse, a production-grade event discovery platform that aggregates 9+ sources into a feed-first experience with provenance metadata, trust tiers, and deterministic deduplication so visitors can rely on what they see.
National Science Foundation (sanitized)
Research proposal triage: SciBERT embeddings + clustering pipeline
Designed a SciBERT embedding and clustering pipeline (HDBSCAN, k-means, Bayesian optimization) that classified 7,000+ research proposals into 70+ themes for reviewer decision support. Ambiguous proposals flagged for human review rather than forced into poor-fit clusters.
National Science Foundation (sanitized)
Panel Wizard: ML-assisted proposal panel formation
Consolidated 8 screens into 1. Sentence embeddings and K-Means clustering suggest proposal groupings, with drag-and-drop overrides so program staff keep full control. Cut panel formation from weeks to hours.
USDA (organic program)
USDA Organic Analytics: Global data warehouse and Tableau reporting for 50,000+ operations
Built a global data warehouse on AWS that brought together Salesforce, an integrity database, and CBP customs records. Added an NLP classifier for organic import taxonomy and dozens of Tableau reports serving 50,000+ certified operations.
VisiTime, LLC
VisiTime: Turning geospatial data into an AR visitor experience
Founded an AR startup, built the tour system on Unity, raised $200K+, shipped a six-hour interactive iPad tour, and earned two U.S. utility patents. Learned things about focus and tradeoffs that consulting never taught me.
How I work
Discover
Sit with the people doing the work. Find the decision, the constraints, and the step where a model genuinely belongs. Requirements documents come after, not before.
Architect
Design the system end-to-end: data pipelines, model workflows, integration points, governance controls, and audit trails.
Prototype
Build working software quickly. Validate assumptions with real users and real data before committing to scale.
Operationalize
Harden pipelines, document everything, establish monitoring. Make it reproducible and maintainable.
Drive adoption
Train users, run stakeholder reviews, retire legacy processes. A tool nobody uses is a tool that failed.
“Professional and collaborative, uniquely suited for DAO work.”Data Analytics Officer, NSF Engineering Directorate
Worth a conversation?
If you need someone to find where AI belongs in a government workflow and carry it through to adoption, I'd be glad to talk.