Find the real workflow
Interview users, inspect artifacts, and surface the tacit rules that do not appear in the request.
Arthur Shafer · Forward-deployed AI / Enterprise systems
Governed AI systems, built end to end.
I turn ambiguous domain problems, fragmented data, and tacit expert workflows into evidence-grounded AI systems that people can actually use.
Active TS/SCI
National-security experience with a bias toward explicit authorization and evidence boundaries.
Sole creator
Rhode Island Political Intelligence System — statewide public evidence, geography, agents, and campaign products.
AWS + Azure
Delivery experience across serverless, containers, data platforms, and enterprise identity.
15 years
Across intelligence, enterprise architecture, technical delivery, and AI product building.
Selected systems
The hard part is rarely the model.
These systems begin with fragmented evidence, hidden workflow rules, regulated users, or an unfamiliar operating environment—and end as software people can interrogate, evaluate, and run.
-
01
Rhode Island Political Intelligence System
A statewide campaign research and strategy workspace connecting public records, elections, geospatial voter-roll intelligence, candidate dossiers, money, public decisions, and an evidence-native agent.
What you'll see: statewide map · representative dossier view · evidence-bounded agent answer
-
02
Governed intelligence platforms
A production federal opportunity-intelligence system evolved into a durable enterprise agent architecture with contracted tools, corrective retrieval, evaluation, and authorization built into the runtime.
-
03
State-government delivery fleet
A repeatable pipeline turns solicitations into working demonstrations, technical packages, and deployment-ready software across corrections, public services, justice, and assessment workflows.
-
04
Theoria
A public, source-aware Orthodox study product built on a retrieval engine, domain-profiled corpus, conversational application, durable threads, and protected edge/origin deployment.
-
05
Product platforms
Serverless products for contractors and professionals: customer provisioning, identity, billing, structured AI document workflows, hosted sites, and usable business software.
How I work
Stay with the problem until the system is operating.
Forward deployment is a continuous loop between domain experts, evidence, product decisions, software, and the constraints of the target environment.
Establish the evidence boundary
Define sources, identity, authority, freshness, claim limits, and the tools the agent is allowed to use.
Build and evaluate
Ship the smallest real workflow, observe failures, add regression and adversarial tests, then widen deliberately.
Deploy and transfer
Package the runtime, identity, observability, runbooks, and decision history so another team can own it.
Enterprise credibility
Technical judgment shaped by operations.
My path runs from Army intelligence through enterprise transformation and aerospace systems to governed AI delivery. The thread is translating consequential work into systems people can act on.
Active TS/SCI clearance · 15 years across intelligence, enterprise architecture, technical delivery, and AI product building.
Booz Allen Hamilton
February 2019 – present
Applied AI delivery
May 2026 – present
Enterprise architecture, technical program leadership, and national-security software delivery.
CVS Health
December 2017 – February 2019
Senior Consultant, Technology and Innovation: modernization, requirements, and business-to-engineering translation.
SpaceX
June 2017 – October 2017
Business Systems Analyst intern: enterprise systems analysis and implementation support in a high-velocity engineering environment.
U.S. Army
January 2011 – June 2014
Intelligence Analyst, 82nd Airborne Division: all-source intelligence, operational planning, and decision support.
Ask about the work
A portfolio you can question.
The assistant retrieves from a versioned set of approved public claims, routes the question, shows the evidence it selected, preserves caveats, and links each answer back to the work.