I share walkthroughs explaining AI governance, explainability, fairness audits, and GDPR-aligned evidence packs on my YouTube channel:
https://www.youtube.com/@suelook9562
Start here: Read the full paper (PDF)
SSRN abstract:View abstract page
This repo is a preview. The PDF is the playbook - it contains the full method, interpretation guidance, and audit-ready evidence structure (not just charts). Written as an evidence pack - not a blog post.
This paper examines the legal, ethical, and technical trade-offs involved in using protected attributes (e.g., sex, race, age, disability) in AI systems. It introduces a governance-led decision framework to support fairness testing while remaining compliant with GDPR and aligned to the EU AI Act.
Who this is for:
AI governance and assurance teams
GDPR Article 22 reviews and special category data governance
EU AI Act high-risk system assessments
Fairness, bias, and explainability audit teams
YouTube walkthrough (now live): https://youtu.be/hgwjOPxe5P4
SSRN paper (now live): https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6020234
- Figures: visuals/
- Evidence pack: publications/
If you work in AI governance, Model Risk Management, Compliance, Risk & Controls, Data Protection, Supplier Assurance, or Responsible AI, the README visuals are only the preview.
The PDF is the full evidence playbook: it shows how to take model outputs (explainability + fairness results) and turn them into audit-ready, governance-ready documentation that supports review, challenge, and accountability in real operational settings.
Because “we have SHAP” is not the same as “we can defend this decision process.”
In practice, governance teams need:
- a clear explanation of what drives outcomes at model level,
- a way to spot and communicate potential disparities across groups,
- language and structure that can survive internal scrutiny, audit, and regulatory questioning,
- a repeatable approach to documenting risk and safeguards where automated decisions have significant impact.
This paper is written to help you do exactly that.
- A repeatable audit logic: how to structure your evidence trail so it’s reviewable and defensible.
- Model-level explainability (SHAP global): how to interpret key drivers and communicate them clearly.
- Fairness testing evidence (sex and age group): how to read selection-rate patterns and flag potential disparity signals.
- Interpretation that leads to action: what to do next (monitoring, thresholds, escalation, documentation updates) when signals appear.
- Governance framing: how to connect technical results to accountability, transparency, and meaningful oversight in practice.
- 2 minutes: open PDF → scan headings and figures to understand the flow.
- 10 minutes: read the explainability and fairness interpretation sections.
- 30 minutes: follow the full evidence chain and reuse the structure for your own model review pack.
If you want the real value - the logic, the interpretation, and the evidence structure you can reuse at work - open the PDF. The README shows the highlights; the paper gives you the full governance method.
This repository is a Responsible AI audit evidence pack showing how to translate:
- Explainability (SHAP = global patterns and case-level explanations)
- Fairness testing (group metrics and disparity interpretation) into GDPR-ready documentation for automated decision systems (Article 22-relevant contexts).
- Working paper (SSRN): https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5783263
- Key visuals:
visuals/(SHAP + fairness outputs) - Evidence pack / publication files:
publications/ - GitHub Pages preview: https://22ifeoma22.github.io/responsible-ai-portfolio-preview/
If you work in AI governance, Model Risk, Compliance, Risk & Controls, Data Protection, Supplier Assurance, or Responsible AI - don’t stop at the charts.
Open the PDF first: publications/ssrn-5783263.pdf This repo is a preview evidence pack - but the PDF is the playbook: how to document, interpret, escalate, and explain results to stakeholders.
- Charts don’t pass audits. The PDF shows how SHAP + fairness outputs become audit-ready evidence.
- Governance = documentation. You get reusable structure + wording for review packs and sign-off.
- It answers “what next?” It includes interpretation guidance + actions when disparities appear.
- 60 seconds: scan the Key Figures in this README (teaser only)
- 5 minutes: PDF → figures + interpretation
- 15 minutes: evidence framing + recommended actions
Bottom line: the README visuals are the hook - the PDF is the toolkit
Best starting point: Open the PDF - the README is a preview, the PDF contains the full method and interpretation + evidence template.
- Explainability (SHAP): what drives outcomes at model level
- Fairness (sex + age): selection-rate disparities + what they mean
- Evidence mapping: how to package outputs into audit-ready documentation
If you do model reviews / governance sign-off, this is the reusable workflow.
.png)
.png)
| Evidence artifact | What it shows | Governance question answered |
|---|---|---|
| SHAP (global) | Main drivers across population | “Can we explain outcomes at model level?” |
| LIME (local) | Why one person got an outcome | “Can we support review/challenge for individuals?” |
| Fairness metrics | Group disparities | “Are outcomes disproportionately adverse?” |
| Audit narrative | limits + safeguards | “What controls reduce risk and enable rights?” |
If you want the extended evidence pack (more artefacts + templates), open an Issue titled “Request: extended pack”.
Title: From Model-Level Explanations to Legal Evidence: Auditing Bias and Explainability in GDPR-Compliant Automated Decision Systems
Read on SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5783263
Article 3 (Now Live) : Ethics Becomes Enforceable Governance When Harm is Foreseeable and Preventable
Read on SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6128487 This paper establishes the operational boundary between ethical intention and enforceable governance. Using a healthcare wearable fall monitoring use case, it demonstrates how foreseeable harm becomes measurable audit evidence through structured monitoring, explainability (SHAP, LIME), logging, escalation, and lifecycle governance.
Aligned with:
- EU AI Act (risk management, monitoring, accountability)
- GDPR (Article 22, fairness, accountability, auditability)
- ISO/IEC 42001 (AI management system lifecycle governance)
- ISO/IEC 27001 and 27701 (security and privacy controls)
Start here: Read the full paper (PDF)
SSRN abstract: Add your SSRN link here