One-line: an 8-minute interactive lesson that teaches a support team to verify AI-drafted customer replies — built as a lesson with governance, not a demo.
Every “AI literacy” portfolio piece I reviewed was either a policy PDF or a prompt-engineering demo. The gap I aimed at: the moment of send. Teams do not fail because they lack AI policy; they fail because a confident draft bypasses the 40 seconds of checking that would have caught it. So this kit teaches a verification behavior (the Verify ladder), wraps it in team guidelines with enforcement teeth (#ai-catch tagging), and ships a facilitator note so a team lead can run it without me in the room.
| File | What it is |
|---|---|
01-lesson-interactive.html | The micro-lesson as a real-time 3D scene (Three.js r160, offline, on the shared v3 scaffold): an agent’s workstation on a contact-centre floor at 17:52, end of the day, with the city at dusk beyond. The AI draft sits in the reply composer on the main screen, word for word as in the panel; the 3-of-5 hotspot hunt marks each real risk up on the screen (passages can be selected on the screen too); the four rungs take the agent to their physical sources in order: the knowledge base on the second screen, the ticket record and the desk-pad calendar, the printed brand-voice card on the divider, and the team lead’s chat. Each cleared rung fixes its line in the on-screen draft and ticks the checklist pinned to the divider. The send decision ends with a single cut to the customer’s phone on his kitchen table, showing the reply that was sent. Rungs: KB match · Ticket math · Brand voice · Owner for money. Hotspots are keyboard-operable and rung options shuffle per visit; no sound. The separate 3d.html is retired and redirects here. Frames from the build are in storyboard.md. |
01-lesson-text.md | The lesson as text (no WebGL needed): the draft, the three risks and two safe passages with their reasons, the four rungs with their evidence and the right fix, and both send outcomes. Generated from the lesson's own content module (lesson-data.js) by tools/build_al_text.mjs, so the two cannot drift. The 3D lesson links it as its text version. |
02-guidelines-onepager.md | Team-level AI-use guideline: 5 rules (each citing the concrete failure it prevents), the Verify ladder table, escalation path. Written in the “richtlijnen” register Dutch teams expect. |
03-facilitator-note.md | How a team lead runs the lesson in a 15-minute huddle: minute-by-minute plan, debrief questions, guardrails, adaptation rules. |
| Role / org type | What this kit demonstrates |
|---|---|
| Lirin-type AI-learning vision roles | AI literacy designed as behavior change with measurement, not content delivery |
| Ricoh-type AI-literacy programs | Scalable micro-learning + facilitation layer + quality signal (#ai-catch volume) |
| UMC-type guideline work | Policy writing in the Dutch “richtlijnen” register with enforcement and escalation paths |
| Exact / SaaS enablement | Support-desk realism: SLA math, KB governance, Billing ownership of money claims |
#ai-catch volume is a process metric; I deliberately do not claim it predicts customer-satisfaction gains without a longer measurement window.Single-session build. Quality gates: node --check on all script blocks, Chromium headless render smoke-test, zero placeholder markers left in any file, exactly one coherent correct path per interaction. AI-assisted drafting disclosed per artifact; all thresholds, war stories and final copy are human-owned.
by I Wayan Usada · runs fully offline · Nordwind Software is a fictional client
STATUS: DELIVERED 2026-08-29 — final artifact published as part of the six-kit portfolio suite (kits 1–6).
action-map.md — Action map — the behaviors the artifact must produce, traced to the business goalstoryboard.md — Text-based storyboard — the scene plan written before buildvisual-mockups.md — Visual mockups & style guide — mood, type, key screens, component decisions