Case study · Independent Portfolio Project
Building an evidence-governed AI operations system. A human-directed, AI-assisted project connecting research, workflow, content operations, documentation, testing, and durable proof.
Prepared by Esse Obayando · AI Operations Specialist
Why the project exists
Disconnected work weakens traceability. Research, source verification, drafting, production planning, review, and quality assurance are often handled as separate activities. The project tests a single operating model that makes boundaries, decisions, evidence, and approval states visible.
Objective: demonstrate an end-to-end system that can turn an ambiguous topic into review-ready professional outputs without bypassing human judgment.
Role and authorship
Direction and execution are not conflated. Project owner Human direction Requirements, project direction, research boundaries, workflow design, architecture approval, quality standards, evidence rules, review, positioning, and final decision authority.
AI assistance Codex implementation Implementation, research support, structured extraction, document generation, automated validation, testing, formatting, and rendering under human direction.
Method
Five views of one controlled system.
End-to-End Project Workflow
How does an idea become review-ready evidence without bypassing human approval? A controlled path from intake through research, content, QA, human review, and evidence preservation.
End-to-End Project Workflow
How does an idea become review-ready evidence without bypassing human approval?
OPERATIONS
Idea / Project Intake
HUMAN AUTHORITY
Scope Definition
RESEARCH
Research Question
RESEARCH
Source Collection
RESEARCH
Source Assessment
RESEARCH
Claims Ledger
RESEARCH
Conflict Review
DECISION / BOUNDARY
Approved-for- Drafting Claims
CONTENT
Content Development
QA / CONTROL
Claim Recheck
OPERATIONS
Production Planning
QA / CONTROL
Quality Assurance
HUMAN AUTHORITY
Human Review
VALIDATED OUTPUT
Publication Preparation
EVIDENCE
Archive / Evidence Preservation
Release Candidate v1.0.0-rc.1 — Evidence Level 4 — Human-directed AI
DGM-001 | Phase 5
Figure 1 End-to-End Project Workflow A gated path from project framing through research, production, quality review, and preserved evidence.
Human-Directed AI Operating Model
Which decisions remain human, and where does AI assist execution? Human authority defines and approves; AI supports bounded execution and returns work for judgment.
Human-Directed AI Operating Model
Which decisions remain human, and where does AI assist execution?
frames
directs and constrains
returns for judgment
produces
retains authority
HUMAN AUTHORITY
HUMAN AUTHORITY Requirements - Scope - Architecture decisions Research boundaries - Quality standards Claim approval - Professional positioning Final approval
AI-ASSISTED EXECUTION
AI-ASSISTED EXECUTION Research assistance - Structured extraction Draft and code generation Document generation - Automated validation Testing - Formatting
EVIDENCE
INPUTS Goals - constraints - evidence
VALIDATED OUTPUT
OUTPUTS Review-ready artifacts and records
DECISION / BOUNDARY
OPERATING RULE AI assists execution. Human authority governs decisions.
Release Candidate v1.0.0-rc.1 — Evidence Level 4 — Human-directed AI
DGM-002 | Phase 5
Figure 2 Human-Directed AI Operating Model Human authority defines scope and approves claims while AI assists bounded implementation and validation.
Phase 4 Research Evidence Map
How did the verified research set narrow into draft-eligible claims and content? Validated internal process counts show how sources narrowed into claims, findings, drafts, QA, and human review.
Phase 4 Research Evidence Map
How did the verified research set narrow into draft-eligible claims and content?
verified
excluded
evidence base
RESEARCH
24 Candidate Sources
VALIDATED OUTPUT
18 Accepted
BLOCKED / NOT CLAIMED
6 Rejected and retained
EVIDENCE
27 Claims Evaluated
VALIDATED OUTPUT
21 Approved for Drafting
BLOCKED / NOT CLAIMED
4 Rejected
DECISION / BOUNDARY
2 Unresolved
RESEARCH
Research Findings
CONTENT
Content Drafts
QA / CONTROL
Quality Assurance
HUMAN AUTHORITY
Human Review
DECISION / BOUNDARY
PROCESS METRICS - NOT COMMERCIAL OUTCOMES
Release Candidate v1.0.0-rc.1 — Evidence Level 4 — Human-directed AI
DGM-006 | Phase 5
Figure 3 Phase 4 Research Evidence Map The locked research snapshot connects 24 reviewed sources to 27 evaluated claims and governed outputs.
Responsible AI Marketing Model
Which marketing capabilities can AI assist, and which safeguards govern trustworthy use? AI-assisted capabilities operate inside human controls for truth, disclosure, fairness, rights, relationships, and approval.
Responsible AI Marketing Model
Which marketing capabilities can AI assist, and which safeguards govern trustworthy use?
grounds
constrains and reviews
assisted work
human judgment
AI-ASSISTED EXECUTION
AI-ASSISTED CAPABILITIES Copy assistance - Virtual staging alternatives Media packaging - Video extension Natural-language search - Lead follow-up Preference-aware content
HUMAN AUTHORITY
HUMAN GOVERNANCE Property accuracy - Disclosure - Fair housing Copyright / rights - Brand judgment Client relationships - Final approval
EVIDENCE
INPUT Verified property and audience context
VALIDATED OUTPUT
OUTPUT Review-ready marketing material
DECISION / BOUNDARY
CORE MESSAGE AI expands workflow capability. Human judgment governs trustworthy use.
BLOCKED / NOT CLAIMED
NO CLAIM OF IMPROVED SALES OUTCOMES
Release Candidate v1.0.0-rc.1 — Evidence Level 4 — Human-directed AI
DGM-007 | Phase 5
Figure 4 Responsible AI Marketing Model AI-assisted capabilities remain bounded by accuracy, rights, fair-housing, brand, and human-approval controls.
Evidence Maturity Model
What has this project demonstrated, and what has it not yet earned? The flagship has tested, inspectable process evidence at Level 4; Level 5 remains explicitly unclaimed.
Evidence Maturity Model
What has this project demonstrated, and what has it not yet earned?
current evidence
limit
PROCESS
LEVEL 1 Claim Only
EVIDENCE
LEVEL 2 Documented
CONTENT
LEVEL 3 Demonstrated
CURRENT TESTED STATE
LEVEL 4 TESTED CURRENT FLAGSHIP
FUTURE / NOT CLAIMED
LEVEL 5 Externally Validated NOT CLAIMED
CURRENT TESTED STATE
LEVEL 4 MEANS Process exists - Workflow executed - Outputs generated Validation passed - Evidence inspectable - Limitations recorded
BLOCKED / NOT CLAIMED
BOUNDARY No third-party validation is claimed.
Release Candidate v1.0.0-rc.1 — Evidence Level 4 — Human-directed AI
DGM-005 | Phase 5
Figure 5 Evidence Maturity Model The project reaches internally tested Level 4 evidence while clearly withholding Level 5.
Research findings
Expanded capability, bounded conclusions. Observed capability AI can assist copy, staging alternatives, media packaging, video extension, natural-language search, lead follow-up, and preference-aware content.
Required controls People retain property-fact verification, creative judgment, rights clearance, disclosure, fair-housing review, relationship management, and final approval.
Outcome boundary No accepted source establishes a universal causal effect on luxury sale price, conversion, time on market, sales volume, or revenue.
Evidence
Tested internally, not validated externally. Level 4 — Tested 24 Sources reviewed 18 Accepted 6 Rejected 27 Claims evaluated 21 Draft eligible 92 Evidence records Level 4 means the workflow was executed, outputs were generated, automated and visual checks were performed, and evidence is inspectable. Level 5 — Externally Validated is not claimed.
Limitations, lessons, next proof
What remains unearned. Limitations Luxury-specific empirical evidence remains limited. The legal review is primarily U.S.-focused, with California as one state example. Vendor sources support feature descriptions, not independent performance outcomes. Product capabilities, policies, and laws can change after the research lock. No client, employer, publication, commercial, or third-party validation is claimed. Lessons Approval state and claim eligibility must be separate. Rejected sources and unresolved claims are useful evidence, not discarded noise. Human authority should be designed into the workflow and visible in the output. Recovery records and version control are part of professional quality. Future improvement The next proof step would be qualified external review or controlled real-world use under a separately approved phase. That step has not occurred.
Case Study Release Candidate
