Find where AI actually matters
Sit with the executives and the experts. Probe with worked examples and prototypes until the rules they hold but cannot state come out.
Arthur Shafer · Enterprise AI
AI initiatives, discovered to delivered.
The hard part of enterprise AI is deciding what deserves to be built, what the evidence can support, and who owns it afterward. I have led all three, and built the systems that prove it.
Active TS/SCI
National-security experience with a bias toward explicit authorization and evidence boundaries.
AI Solution Architect
Currently, at Booz Allen Hamilton.
15 years
Across intelligence, enterprise architecture, technical delivery, and AI product building.
Live software
Products, systems, and demonstrations running behind this domain.
Selected systems
The hard part is rarely the model.
The model is the last one percent of the effort. The other ninety-nine is the data underneath it, the process it has to serve, and the judgment about what AI could do here, and what it should. Each of these six systems begins with that work.
-
01Case study
Rhode Island Political Intelligence System
Statewide campaign research over public records, geography, money, and votes, with an agent that shows its evidence and its limits.
-
02Live: /contracts ↗
Governed intelligence platforms
A federal opportunity system evolved into an enterprise agent platform: contracted tools, corrective retrieval, and authorization inside the runtime.
-
03The demonstrations
State-government delivery fleet
Solicitations turned into working demonstrations across corrections, justice, public transactions, and scheduling.
-
04Live: /theoria ↗
Theoria
A public, source-aware study product with durable threads and a protected origin.
-
05Both products
Product platforms
Contractor CRM and ResFol.io: serverless products from customer event to hosted experience.
-
06Case study
Oracle
An autonomous research and signal system: asset maps, research agents, generated collectors, a validation gauntlet, and daily refresh.
How I operate
Stay with the initiative until it runs without me.
Inside an organization, an AI initiative is a sequence of decisions before it is a system. This is the sequence I run.
Decide what to build, and what not to
When leadership disagrees, sequence instead of picking a side: the shared data first, then the products that need it, in parallel.
Build the smallest thing that can be measured
Evidence contracts, an evaluation harness as the gate, and a cost ledger, so every change has a score and a price.
Leave it running in someone else's environment
A deployment contract, a restore runbook, and a team staffed around it. The work is done when it no longer needs me.
Live
Real software, running.
Products, systems, and demonstrations, each labelled for what it is. Nothing here is a mockup.
Product · live
Theoria
Source-aware scripture study with durable threads.
/theoria ↗System · live
Federal opportunity intelligence
The live predecessor of the enterprise agent platform.
/contracts ↗System · authenticated
Oracle Control Room
Read-only view of the autonomous research and signal system.
Case study →Demonstrations · running
State-government delivery fleet
RI assessor, TN vehicles, KY public advocacy, Maine booking.
Open the fleet →Demonstration · live
Corrections voice interview
Adaptive voice and interview orchestration.
/ridocdemo ↗Product · live
ResFol.io
A résumé becomes a hosted professional site.
resfol.io ↗Writing
How I think about an organization's AI problem.
A curated set of case notes: the situation, the complication, the decision, what it cost, and the lesson small enough to be true.
-
Case note
Deterministic scaffolding, or let the model work
Read this if you are deciding where the code stops and the model starts.
-
Case note
Tiered data access and the answer you cannot give
Read this if two users with different clearances will ask the same system the same question.
-
Case note
Sprawl in agentic coding
Read this if an agent-built feature keeps growing and nobody asked it to.
Experience
Technical judgment shaped by operations.
From Army intelligence through enterprise transformation to governed AI delivery. The thread is turning consequential work into systems people can act on.
MBA and Bachelor of Science, Bryant University
Booz Allen Hamilton
February 2019 – present
Artificial Intelligence Solution Architect
2026 – present
Enterprise Architect / Technical Program Manager, 2019 – 2025: requirements decomposition for $10M+ DoD systems; briefed O-5/O-6 leadership.
CVS Health
December 2017 – February 2019
Senior Consultant, Technology and Innovation: modernization portfolio prioritized against measurable ROI.
SpaceX
June 2017 – October 2017
Business Systems Analyst intern: enterprise ERP features and implementation support.
U.S. Army, 82nd Airborne
January 2011 – June 2014
Intelligence Analyst, 1-508th Parachute Infantry Regiment: all-source intelligence and decision support.
Capability
Proven by
Discovery with executives and experts
Enterprise platform · Government delivery
Deciding what to build, and in what order
Enterprise platform
Evidence and authorization architecture
Governed intelligence · Rhode Island
Entity resolution without shared keys
Enterprise platform · Rhode Island
Agent runtime and evaluation loops
/contracts · Theoria · this site's assistant
Enterprise cloud delivery and cutover
AWS serverless · Azure, Entra ID, Databricks
Data fabric design at lake scale
Enterprise platform
Autonomous multi-agent operations
Oracle
Ask
A portfolio you can question.
The assistant answers only from a versioned set of approved claims, shows the evidence it selected, and says when the record does not establish an answer.