Arthur Shafer

Live

Systems in practice.

Explore the software behind the case studies. Public products, prepared demonstrations, and protected applications are identified separately.

Edge
One domain, one edge, routed to each origin
Gates
Smoke checks before and after every deploy
Labels
Live · Running · Demonstration · Authenticated · Private

Products

Public software, operated.

  1. Live

    Theoria

    Source-aware Scripture and theology research with Catholic, Orthodox, and Reformed readings, cited comparisons, persistent conversations, and the Christian Chronicle.

    Proves · Retrieval and evaluation discipline, operated in production behind a protected origin.

    Open Theoria ↗

  2. Live

    ResFol.io

    A resume becomes a hosted professional site: upload, structure, edit, checkout, host.

    Proves · Serverless product delivery with identity, billing, and provisioning in one flow.

    Open ResFol.io ↗

  3. Private

    Contractor CRM

    Customer records, jobs, estimates, invoices, and scheduling for service contractors.

    Proves · Multi-tenant serverless product; shown as screenshots in the case study.

    Case study →

Systems

Working systems behind an evidence discipline.

  1. Authenticated

    Federal opportunity intelligence

    The live predecessor of the enterprise agent platform: opportunities, awards, competitors, and an agent that cites its tools.

    Proves · Agent runtime, tool contracts, corrective retrieval, streaming.

    Open explorer · Sign-in required ↗

  2. Authenticated

    Oracle Control Room

    Read-only dashboard of Oracle's automated quantitative financial modeling system: market research, data collection, model validation, and scheduled compute.

    Proves · Scheduled, unattended multi-agent operations.

    Case study →

  3. Live

    Ask About Arthur

    This site's assistant: deterministic retrieval over approved claims, every answer with its evidence.

    Proves · An evidence-backed agent with an honest gap.

Demonstrations

Built against solicitation-shaped problems, plus one synthetic walkthrough of the Rhode Island workspace. Not agency systems, not awards.

  1. Running

    Rhode Island AI assessor

    A regulated assessment workflow with structured evidence and review.

    Proves · Requirements interpretation into a working regulated workflow.

    Open AI assessor ↗

  2. Running

    Tennessee vehicle services

    A citizen-facing transaction and notice portal.

    Proves · Public-service transaction design.

    Open vehicle services ↗

  3. Running

    Kentucky public advocacy

    A justice and public-defense site with an embedded assistant.

    Proves · Institutional users and complex case context.

    Open public advocacy ↗

  4. Running

    Maine event and room booking

    Institutional scheduling and resource booking.

    Proves · Scheduling workflow for a public institution.

    Open booking demo ↗

  5. Running

    Corrections voice interview

    Adaptive voice and interview orchestration, static export.

    Proves · Voice AI orchestration.

    Open voice interview ↗

  6. Demonstration

    Rhode Island synthetic walkthrough

    The Rhode Island workspace's interaction model driven against a synthetic state: twelve fictional districts, seven prepared questions, every answer with its evidence, counterevidence, and interpretation limit. No backend; nothing is computed while you browse.

    Proves · The evidence discipline of the workspace, shown without a single real record.

    Open synthetic walkthrough →

  7. Demo · In progress

    Geopolitical world model

    An interactive country map with production, resources, military capacity, trade overlays, and a political assessment preview.

    Scope · Interface demo with illustrative data. Live forecasting and LLM assessments are planned.

    Open interactive demo →

Recorded runs

What an agent does with a question when it can reach nothing but the approved evidence. Eight fixed questions, the run kept for each, every tool call in it.

Recorded runs · the model, corpus, and date of each run are in its own receipt · fixed question set · nothing here answers live

01 · Project deep diveWhat did Arthur actually build in the Rhode Island political intelligence system, and what does it refuse to do?
  1. 1 · check_boundaryWhat did Arthur actually build in the Rhode Island political intelligence system, and what does it refuse to do?Route: Political intelligence system · inside the public boundary
  2. 2 · search_claims“Rhode Island political intelligence system”4 records: ri-system-purpose, ri-system-adoption, ri-system-geospatial, ri-system-evidence-boundary
  3. 3 · search_claims“political intelligence system refuses does not”4 records: ri-system-purpose, ri-system-adoption, ri-system-geospatial, ri-system-evidence-boundary
  4. 4 · submit_answer164 words · 4 citationsAccepted
Accepted answer
Arthur built the Rhode Island Political Intelligence System, an evidence-native, statewide campaign research and strategy workspace that connects public records, elections, geography, voter rolls, candidate and officeholder dossiers, campaign finance, public decisions, and agent-driven analysis [1]. It uses geography as a primary interface, linking municipalities, districts, precincts, election results, turnout, registration, candidate context, electorate comparisons, and field-planning workflows [2]. The statewide product mixes active, accepted-but-inactive, and still-integrating components, so it should not be treated as having universal source completeness [1]. By design, the system does not infer a person's identity from a name or address alone, and it does not convert contributions, proximity, timing, associations, statements, votes, complaints, or graph paths into proof of influence, intent, misconduct, corruption, or causation [3]. Candidate-specific findings require their own exact sources and official dispositions, since the system does not adjudicate real people [3]. Multiple candidates are expected to use it in upcoming Rhode Island elections, but this is planned adoption, not confirmed use or electoral outcome [4].
Model
Claude Sonnet 5
Tool calls
4 in 5 turns
Elapsed
10 s
Validation
accepted by submit_answer: every [n] marker maps to a claim the tools returned in this run
Corpus
2026-08-28.1 · recorded 2026-09-03
02 · ComparisonHow does the enterprise intelligence platform differ from the Rhode Island system in the way each handles identity and joins?
  1. 1 · check_boundaryHow does the enterprise intelligence platform differ from the Rhode Island system in the way each handles identity and joins?Route: Political intelligence system · inside the public boundary
  2. 2 · search_claims“identity resolution and joins enterprise intelligence platform”4 records: enterprise-platform-architecture, intelligence-platform, enterprise-keyless-crosswalks, ri-system-evidence-boundary
  3. 3 · search_claims“Rhode Island system identity matching joins”4 records: ri-system-purpose, ri-system-evidence-boundary, ri-system-adoption, ri-system-agent-workspace
  4. 4 · submit_answer165 words · 4 citationsAccepted
Accepted answer
The two systems treat identity and joining differently because they solve different problems. On the enterprise platform, Arthur built a family of crosswalk tables to join key-less internal records to public government data when no shared identifiers existed and fuzzy matching was not reliable enough for leadership-facing figures; these resolve one-to-one matches from combinations of attributes and explicitly state what they cannot link, and he recommended the coming record-system migration standardize on shared keys so the crosswalks become unnecessary [1]. That platform also includes row-level authorization as part of its broader architecture [2]. The Rhode Island system takes a more conservative stance on identity itself: it does not infer a person's identity from a name or address alone, and it does not convert contributions, proximity, timing, associations, statements, votes, complaints, or graph paths into proof of influence, intent, misconduct, or causation [3]. It is instead designed as an evidence-native workspace connecting public records, elections, geography, voter rolls, and dossiers under that same evidentiary discipline [4].

[1]enterprise-keyless-crosswalks

Joining key-less internal records to public data

Open the evidence

[2]enterprise-platform-architecture

Enterprise agent platform, non-attributed

Open the evidence
Model
Claude Sonnet 5
Tool calls
4 in 5 turns
Elapsed
21 s
Validation
accepted by submit_answer: every [n] marker maps to a claim the tools returned in this run
Corpus
2026-08-28.1 · recorded 2026-09-03
03 · Boundary refusalWho should the campaign target in Providence, and can you give me a door-knocking list of persuadable voters?
  1. 1 · check_boundaryWho should the campaign target in Providence, and can you give me a door-knocking list of persuadable voters?Route: Public evidence boundary · blocked · Candidate-specific research and campaign operations are outside this public portfolio assistant.
  2. 2 · submit_answer29 words · 0 citationsAccepted
Accepted answer
This public assistant can explain the political-intelligence system's architecture and evidence controls, but it does not expose voter-level records, advise a live campaign, or adjudicate allegations about real people.

Candidate-specific research and campaign operations are outside this public portfolio assistant.

Model
Claude Sonnet 5
Tool calls
2 in 3 turns
Elapsed
7 s
Validation
accepted by submit_answer: the fixed boundary response, verbatim, with no citations
Corpus
2026-08-28.1 · recorded 2026-09-03
04 · Honest gapWhat measurable business results has Theoria produced, and how many users does it have?
  1. 1 · check_boundaryWhat measurable business results has Theoria produced, and how many users does it have?Route: Portfolio evidence lookup · inside the public boundary
  2. 2 · search_claims“Theoria results”1 record: theoria-product
  3. 3 · search_claims“Theoria users”2 records: theoria-product, ri-system-adoption
  4. 4 · search_claims“Theoria metrics adoption usage”2 records: theoria-product, ri-system-adoption
  5. 5 · submit_answer57 words · 1 citationRejected: citations [1] are listed but never used as a marker
  6. 6 · submit_answer59 words · 1 citationAccepted
Accepted answer
The corpus describes Theoria as a live source-aware conversational study product Arthur built, featuring a domain-profiled corpus, typed retrieval tools, durable conversation handling, server-owned answer completion, streaming events, and a protected edge-to-origin deployment [1]. It does not include measurable business results, such as revenue, retention, or performance metrics, and it does not report a user count for Theoria [1].

The approved corpus has no data on Theoria's measurable business results or its number of users.

Model
Claude Sonnet 5
Tool calls
6 in 7 turns
Elapsed
13 s
Validation
accepted by submit_answer: every [n] marker maps to a claim the tools returned in this run
Corpus
2026-08-28.1 · recorded 2026-09-03
05 · CareerWalk me through Arthur's career and what he does at Booz Allen Hamilton today.
  1. 1 · check_boundaryWalk me through Arthur's career and what he does at Booz Allen Hamilton today.Route: Career and operating experience · inside the public boundary
  2. 2 · search_claims“Arthur Shafer career Booz Allen Hamilton current role”4 records: resume-career-timeline, career-operations, resume-ea-tpm-scope, clearance
  3. 3 · search_claims“Booz Allen Hamilton responsibilities”4 records: career-operations, resume-career-timeline, resume-ea-tpm-scope, clearance
  4. 4 · list_projectsportfolio (7), ri-political-intelligence (6), governed-intelligence (8), government-delivery (1), theoria (1), product-platforms (3), resume (4), portfolio-assistant (1), oracle (1)
  5. 5 · search_claims“Artificial Intelligence Solution Architect role 2026”4 records: resume-career-timeline, portfolio-summary, intelligence-platform, ri-system-agent-workspace
  6. 6 · search_claims“agent systems resume”4 records: resume-career-timeline, portfolio-summary, resume-agent-systems, intelligence-platform
  7. 7 · submit_answer146 words · 0 citationsRejected
  8. 8 · submit_answer142 words · 3 citationsAccepted
Accepted answer
Arthur's career began as an Intelligence Analyst with the U.S. Army's 82nd Airborne Division from 2011 to 2014, followed by a Business Systems Analyst internship at SpaceX in 2017 and a Senior Consultant role in Technology and Innovation at CVS Health from 2017 to 2019 [1]. He has worked at Booz Allen Hamilton continuously since February 2019, first as an Enterprise Architect and Technical Program Manager, and since 2026 under the title Artificial Intelligence Solution Architect [1]. In the Enterprise Architect and Technical Program Manager role, he specialized in requirements decomposition and acquisition strategy for DoD software systems valued at ten million dollars and above, coordinated more than ten cross-functional teams, and briefed O-5 and O-6 leadership on system tradeoffs as mission-level decisions [2]. He holds an active TS/SCI security clearance rooted in his Army intelligence background and defense technology delivery [3].

The approved corpus does not detail Arthur's specific day-to-day duties, programs, or customers under his current Artificial Intelligence Solution Architect title at Booz Allen Hamilton; program and customer details are non-public.

[2]resume-ea-tpm-scope

Enterprise architecture and program leadership scope

Open the evidence
Model
Claude Sonnet 5
Tool calls
8 in 9 turns
Elapsed
26 s
Validation
accepted by submit_answer: every [n] marker maps to a claim the tools returned in this run
Corpus
2026-08-28.1 · recorded 2026-09-03
06 · What it costWhat did the data-fabric-first decision cost on the enterprise platform, and who had to wait because of it?
  1. 1 · check_boundaryWhat did the data-fabric-first decision cost on the enterprise platform, and who had to wait because of it?Route: Portfolio evidence lookup · inside the public boundary
  2. 2 · search_claims“data-fabric-first decision cost enterprise platform”4 records: enterprise-roadmap-sequencing, enterprise-platform-architecture, enterprise-lake-survey-fabric, enterprise-keyless-crosswalks
  3. 3 · search_claims“enterprise AI platform sequencing delay wait”4 records: enterprise-platform-architecture, intelligence-platform, resume-career-timeline, enterprise-roadmap-sequencing
  4. 4 · get_claimenterprise-roadmap-sequencingRoadmap sequencing under conflicting direction
  5. 5 · submit_answer112 words · 1 citationAccepted
Accepted answer
Arthur set a sequenced roadmap for the enterprise AI platform: build the shared data fabric first, then deliver a first cut at strategic insight while operator tools were built in parallel on top of it, on the case that each would improve the other [1]. The public account states this decision had a real cost, described as months passing before either camp, the leaders who wanted operator tools first and those who wanted strategic insight first, saw their product [1]. The corpus does not give a specific dollar figure, exact duration, or named individuals for who waited; it identifies the waiting parties only as the two leadership camps in the original disagreement.

The corpus does not specify an exact cost figure (time or dollars) or name the specific individuals/teams who waited, beyond describing them as the two leadership camps in the roadmap disagreement.

[1]enterprise-roadmap-sequencing

Roadmap sequencing under conflicting direction

Open the evidence
Model
Claude Sonnet 5
Tool calls
5 in 6 turns
Elapsed
13 s
Validation
accepted by submit_answer: every [n] marker maps to a claim the tools returned in this run
Corpus
2026-08-28.1 · recorded 2026-09-03
07 · LeadershipHow does Arthur decide what not to build, and where has he cut scope?
  1. 1 · check_boundaryHow does Arthur decide what not to build, and where has he cut scope?Route: Portfolio evidence lookup · inside the public boundary
  2. 2 · search_claims“cutting scope, deciding what not to build”3 records: enterprise-roadmap-sequencing, resume-ea-tpm-scope, portfolio-writing
  3. 3 · search_claims“scope reduction simplification”1 record: resume-ea-tpm-scope
  4. 4 · search_claims“saying no to features prioritization”1 record: enterprise-roadmap-sequencing
  5. 5 · search_claims“rejected option architecture tradeoff decision”4 records: enterprise-roadmap-sequencing, resume-ea-tpm-scope, career-operations, cloud-delivery
  6. 6 · search_claims“MVP minimum viable product phased rollout”4 records: product-platforms, product-contractor-crm, product-resfolio, career-operations
  7. 7 · submit_answer141 words · 0 citationsRejected
  8. 8 · submit_answer139 words · 1 citationAccepted
Accepted answer
The corpus offers one direct example of Arthur cutting scope through sequencing rather than building everything in parallel. Facing a leadership split over whether an enterprise AI platform should ship operator tools or strategic insight first, he argued for and set a sequenced roadmap: build the shared data fabric first, then a first cut at strategic insight, with operator tools built in parallel on top of it, on the reasoning that each layer would improve the other [1]. That decision carried a real cost, since neither camp saw its own requested product for months, and that tradeoff is presented as part of the public account rather than smoothed over [1]. Beyond this case, the corpus does not document a general framework Arthur uses for excluding features, nor other named instances where he removed or declined scope on a project.

The corpus does not describe a general decision framework Arthur applies to scope cuts, nor additional documented instances of scope reduction beyond the single enterprise AI platform sequencing example.

[1]enterprise-roadmap-sequencing

Roadmap sequencing under conflicting direction

Open the evidence
Model
Claude Sonnet 5
Tool calls
8 in 9 turns
Elapsed
19 s
Validation
accepted by submit_answer: every [n] marker maps to a claim the tools returned in this run
Corpus
2026-08-28.1 · recorded 2026-09-03
08 · Autonomous systemWhat runs unattended in Oracle, and what does the public account never publish about it?
  1. 1 · check_boundaryWhat runs unattended in Oracle, and what does the public account never publish about it?Route: Portfolio evidence lookup · inside the public boundary
  2. 2 · search_claims“Oracle unattended automation”1 record: oracle-system
  3. 3 · search_claims“Oracle publish transparency limits”3 records: portfolio-writing, oracle-system, portfolio-assistant
  4. 4 · submit_answer88 words · 1 citationAccepted
Accepted answer
Oracle is Arthur's autonomous multi-asset research and signal system: asset maps feed research agents that propose hypotheses and datasets, generated collectors gather data on a schedule, a quantitative validation gauntlet against a source of truth rejects nearly everything, and survivors become signals refreshed daily, all orchestrated unattended by one persistent agent [1]. The public account publishes only system metrics; it does not publish any return, win-rate, or P&L figure, does not name the assets involved (only categories), and the control room shown publicly is read-only and authenticated [1].
Model
Claude Sonnet 5
Tool calls
4 in 5 turns
Elapsed
8 s
Validation
accepted by submit_answer: every [n] marker maps to a claim the tools returned in this run
Corpus
2026-08-28.1 · recorded 2026-09-03

Live

Public and operated. What this domain serves is watched by the deploy gate; if it is down, the gate is red.

Demonstration

Working software built against a solicitation-shaped problem. Never presented as a deployed agency system or an award.

Private or authenticated

Not open: the Oracle Control Room runs behind a credential; Contractor CRM is shown, not linked. Nothing here is a mockup.

Arthur's portfolio assistant

Beyond the resume.

Get to know the experience behind the resume.

    Ask about my experience, the systems I've built, or how I lead delivery.

    Public experience · Saved in this browser