CASE STUDY — NSF · 2024–2025

NSF AI Governance Foundations

Building the operating layer for responsible AI adoption inside a newly formed Chief AI Officer function

As GS-15 Lead Data Scientist / AI Oversight advisor reporting directly to NSF's Chief AI Officer, I helped build greenfield AI governance, technical-review, community, security, and data-architecture foundations for responsible AI adoption across a major U.S. science agency.

This work occurred before NSF's later public AI Strategy was drafted and published. I did not author that strategy. My contribution was earlier-stage: building the operating structures, review patterns, community forums, technical architecture, and interagency coordination that helped make responsible AI adoption actionable.

GS-15
Lead Data Scientist / AI Oversight
100+
AI CoP members co-chaired
7
Governance & architecture workstreams

§ 01 — Program contextContext

NSF's Chief AI Officer function was newly forming during a period of rapid change in federal AI policy. The agency had deep AI expertise across its research mission, but internal AI adoption needed an operating model: intake, risk classification, technical review, approval paths, security exercises, data architecture, and community coordination.

The challenge was greenfield: there was no mature AI evaluation process, no established AI governance workflow, and no unified technical-review pattern for internal AI systems.

§ 02 — My role and responsibilitiesMy mandate

I reported directly to the CAIO and worked across technical architecture, governance implementation, interagency liaison, and community enablement.

My responsibilities included AI use-case intake, risk classification, technical-review design, AI deployment guidance, AI Community of Practice leadership, engineering review board participation, security table-top exercises, Microsoft Copilot rollout support, and architecture for a production vector/graph capability.

§ 03 — Workstreams and deliverablesWhat I built and led

01 AI Use-Case Intake and Risk Classification Governance

Owned and improved NSF's AI use-case intake process. Created a risk-classification approach to help distinguish low-risk experimentation from use cases requiring deeper review, documentation, security analysis, access-control review, or governance oversight.

02 Technical Review and Approval Pathways Engineering

Created AI technical-review patterns and served as a voting engineering review board member for AI-related systems. Reviewed AI aspects of proposed systems, including architecture, data flows, access controls, security implications, tooling choices, and production readiness.

03 AI Deployment Playbook Guidance

Created an AI deployment playbook to help teams move from experimentation toward governed implementation. The playbook translated AI governance into practical delivery questions: data sensitivity, human review, transparency, audit readiness, security controls, model behavior, and operational ownership.

04 AI Community of Practice Community

Co-chaired a 100+ member AI Community of Practice spanning technical, policy, security, mission, and leadership functions. The community served as a discussion forum, coordination layer, and announcement channel for AI rollouts, including Microsoft Copilot-related enablement.

05 AI Security Table-Top Exercises Security

Ran AI security table-top exercises to help stakeholders reason through AI-specific risks before deployment, including data exposure, prompt injection, tool access, misuse scenarios, and governance gaps.

06 Vector and Graph Architecture for Research-Impact Intelligence Architecture

Served as technical architect for a production vector/graph capability designed to help NSF understand the network effects and downstream impact of funded research. The system used Amazon Neptune for graph relationships, embeddings generated through AWS Lambda, and a user interface designed to mimic existing workflows so staff could adopt it without changing how they worked.

07 Interagency Liaison Coordination

Liaised with NASA, DOE, NAIRR stakeholders, and other federal partners on emerging AI governance and adoption patterns. Work included comparing Microsoft Copilot rollout approaches with NASA and discussing standardized data-tagging practices with DOE.

§ 04 — Practical resultsOutcomes

  • Created repeatable AI use-case intake and risk-classification patterns.
  • Connected AI governance to engineering review and production approval.
  • Built AI deployment guidance for internal teams.
  • Co-chaired a 100+ member AI Community of Practice.
  • Supported Microsoft Copilot rollout governance and enablement.
  • Ran AI security table-top exercises.
  • Architected a production vector/graph capability for research-impact intelligence.
  • Brought lessons from NASA, DOE, NAIRR, and other federal partners into NSF's internal AI adoption work.

§ 05 — TransferabilityWhy it matters

Many organizations do not fail at AI because they lack prototypes. They fail because they lack the operating layer around prototypes: governance, intake, technical review, data architecture, evaluation, security, funding pathways, user adoption, and executive alignment.

This work built that operating layer in a high-scrutiny federal environment. The experience transfers directly to regulated enterprise GenAI, AI assurance, model governance, public-sector AI, financial-services AI risk, and responsible deployment of LLM systems.

Timing note

This work occurred during the early build-out of NSF's Chief AI Officer function and before NSF's later public AI Strategy was drafted and published. I did not author the published strategy. My contribution was earlier-stage: building governance, review, community, security, interagency, and data-architecture foundations that preceded later agency-wide strategy work.

← Back to profile

AI Governance · Federal AI · GS-15 · C·C ◆ Last updated May 2026