AI Champions Interactive Labs
CORNERSTONE · Week 1 · AI Foundations · Capstone

Cornerstone: The Cornerstone Project — Four Desks, One Afternoon

Everything you need to run today is on this page and inside your desk — no instructor walkthrough required. Read this once, then go.

Every desk below is one of the "five moments" from your Day 4 prompt library — now played out as a full working day. Your group gets one desk. Everything you need — the fictional dataset, the brief, and every deliverable — lives inside that one lab page. Nothing here is real Burgan data; nothing you build here should ever be pointed at a real Burgan system.

Start here — before you open a desk

  1. Get into groups of 4. More than 16 people in the room? Double up — two groups can run the same desk independently and compare notes at presentation time.
  2. Agree on one desk. If two groups want the same one, whoever decides second picks a different one.
  3. Check your tools before you start the clock: everyone should have Claude open in a browser tab (same as every day this week), and at least one person in your group should have a free account at gamma.app ready — you'll need it later for your presentation, and it's free to sign up.
  4. Pick one "scribe" — whoever's laptop holds your group's official answers in the lab page. Everyone else still works in parallel (their own Claude tabs, notes, or scratch paper) — see the "How to split this" tip under every tab inside your desk for exactly how to divide the work so nobody's idle.
  5. Open your desk below, read the brief, download the dataset, and start with Triage. That's the whole kickoff.
Not presenting yet? While another group presents, open Audience Notes on your own device — jot one thing you'd steal and one question for each desk as you watch. You'll share it during that team's Q&A.
Today's shape (~5 hours of real work): triage the desk's tasks → run a safety/data-classification pass → build a small prompt library → design and run an agent across the whole dataset → validate the output against the source → write your Agent Card → present your desk in a short Gamma.app deck. The model answers in seconds — the five hours are almost entirely your team thinking, checking, and building, not waiting on AI.
C1 Sunday night — drafting Monday's escalation note

The AML Escalation Desk

Eighteen flagged alerts. One Monday-morning deadline. Build the process, not just the notes.

Open this desk →
C2 Monday 9am — a customer complaint just hit your inbox

The Complaint Handover Desk

Sixteen overnight complaints. One morning to triage, draft, and hand each one off safely.

Open this desk →
C3 Tuesday lunch — three credit reviews waiting for committee

The Credit Committee Desk

Eight loan files. One committee meeting. The number that matters better be checked.

Open this desk →
C4 Friday 4pm — weekly overdue-loan reminders

The Overdue-Loan Batch Desk

25 accounts. One afternoon. One instruction that has to work every single time.

Open this desk →

Every group's deliverables — same seven, different desk

  1. Triage — classify this desk's recurring tasks by shape and risk.
  2. Safety pass — decide what belongs in a prompt and what doesn't, even on fictional data.
  3. Prompt library — 2-3+ reusable, iterated prompts for this desk's repeating task.
  4. Agent design — a trigger, instructions, and a guardrail, actually run across the whole dataset.
  5. Validation log — cross-check the agent's output against the source. Something in every dataset doesn't add up — find it.
  6. Agent Card — what this agent may do alone, what always needs a human, and what it would take before this pattern could touch a real Burgan system.
  7. Presentation — a short Gamma.app deck, presented live to the room.
The doorway rule, still in force: if your group builds the live connector stretch goal, it runs on a personal account against the downloaded fictional dataset — never a Burgan account, never a real customer record.