01 Evidence synthesis
Defender UX Signal Agent
Defined how messy UX evidence becomes decision records, ADO-ready UX issues, validation plans and PM/engineering summaries — separating evidence from interpretation.
← Back to work
AI · Production workflow
AI Security Experiences
From static artifacts to an AI-assisted production workflow. My role shaping the NPF workstream for Microsoft Defender UX — reusable building blocks, SFE- and Fluent-backed prototypes, accessibility gates, signal synthesis, and an emerging operating model for how Defender UX moves faster with evidence and control.
ProblemAI work was arriving faster than the design system could absorb it.
Defender UX wasn't missing ideas. It was missing a repeatable way to turn scattered signals into implementation-ready product work: evidence, decisions, prototype behavior, accessibility requirements, SFE/Fluent constraints, and ADO-ready summaries.
The opportunity was to treat AI not as a chatbot layer, but as a production capability for the design organization — helping designers synthesize messy inputs, prototype faster, preserve human control, and create reusable patterns engineering could trust.
- Signals were spread across Teams, Outlook, meetings, Loop/SharePoint, Figma, ADO, prototypes and local repos.
- SFE and Fluent needed to move from reference material into reusable, implementation-facing building blocks.
- Accessibility needed to be part of production, not late-stage cleanup.
- AI-assisted navigation had to stay transparent, reversible, and governed by role, permission and confidence signals.
My contributionBuilding blocks for a new UX production function.
02 Prototype operating model
Prototype Lab
Moved beyond mockups into coded prototypes, reusable prompts, local repos, pattern audits and video storytelling teams could inspect and reuse.
03 Design-system bridge
SFE + Fluent adoption
Connected Figma, SFE, Fluent 2, page templates, component constraints and quality gates so design intent could survive implementation.
04 Trust model
Quantum UX for AI
A responsible interaction model for ambiguous intent, confidence, transparency, control and reversibility in AI-assisted navigation.
05 Accessibility by design
A11y production gates
Accessibility scorecards, WCAG-oriented issues, semantic headings and review readiness built into the workflow before handoff.
06 Delivery language
ADO-ready outputs
Turned ambiguity into structured summaries, acceptance criteria, validation metrics, SFE notes and clear next actions.
ProgressionFrom prototype foundation to a reusable production model.
April
Prototype foundation
Connected Edge, Figma, SFE, Copilot, ADO, GitHub, Teams, Outlook and local Defender prototypes into one working production surface.
May
System adoption
Shifted toward SFE components, Figma kits, Fluent-backed implementation, accessibility checks and Prototype Lab workflows.
June
Proof point
A Copilot-powered Defender navigation personalization prototype using role, JTBD, usage, confidence and policy signals to shape pinned navigation.
Next
Emerging baseline
Signals become specs, specs become SFE-grounded prototypes, prototypes become decisions, and decisions become ADO-ready outputs.
WalkthroughThe contribution story, start to finish.
Case exampleDefender navigation onboarding made the future tangible.
The clearest proof point: a running Defender prototype that changes onboarding behavior through role, JTBD, usage and hybrid navigation recommendations — inspectable in the real product shell, not a static mockup.
Design decisionsNot "decide for the user" — suggest, explain, and let the user steer.
- Grounded by default. Every recommendation needs traceable evidence and a confidence signal.
- Human in control. AI proposes and stages changes; the analyst or admin confirms consequential decisions.
- Legible automation. The system makes visible what changed, why it changed, and how to undo it.
- Reusable by design. Patterns map to SFE, Fluent, accessibility and engineering handoff constraints.
- Production-ready outputs. The endpoint isn't a prettier mockup — it's a reviewed decision, validation plan and ADO-ready work item.
Outcome & learningA repeatable pattern language for AI-assisted product work.
FasterPrototype cycles moved closer to working product behavior instead of static slide decisions
TrustedGrounding, confidence, reversibility and A11y became part of the product logic
ReusableSFE, Fluent, templates, prompts and repos became building blocks for future work
ActionableEvidence could convert into summaries, validation plans and ADO-ready UX issues
The contribution worth claiming: Defender UX moved from static artifacts toward a repeatable AI-assisted production workflow.
Not just that I designed an AI prototype — but that I helped define how AI, SFE, Fluent, accessibility, prototype labs and evidence synthesis can become the operating model for future product work.
Evidence basisGrounded in real artifacts and work signals.
| Evidence | What it supports | Source type |
|---|---|---|
| NPF narration & portfolio narration | April–June progression, NPF framing, production-function thesis, Signal Agent and Prototype Lab story. | Local project files |
| Defender navigation onboarding prototype & share-out | Role / JTBD / usage / confidence / policy-based personalized navigation and prototype proof point. | Local images + M365 metadata |
| Defender UX Signal Agent backlog & prompts | Evidence-to-decision workflow, ADO-ready UX issue templates, PM/engineering summaries, SFE/prototype/A11y handoff fields. | Local repository files |
| SFE / Fluent / page-template artifacts | Design-system adoption, page-list recommendations, implementation-facing patterns and Fluent-backed constraints. | OneDrive metadata + local decks |
| A11y tracking & prototype accessibility strategy | Accessibility as production gate: scorecards, annotations, semantic structure and WCAG-oriented issues. | M365 metadata + local workflow files |
| Prototype Lab & video storytelling | Operating model for reusable coded prototypes, prompts, repos, pattern audits and design-to-engineering translation. | Local video/image artifacts |