Skip to content
The AI Scale Academy

Guided lab · simulated

Build an inbox triage workflow — safely.

Eight made-up customer messages. Write keyword rules to route each one. Anything your rules don't catch goes to a person, which is the safe default. No real email, no AI and no data leaves this page.

1 · Your rules

The first rule that matches wins, so order matters. Course reference: Triage, reminders and safe escalation.

2 · Results

Add rules, then run the simulation.

Members' labs · simulated · practice only

Permission-safe retrieval: test the evidence boundary

Configure a retrieval policy for a fictional support assistant and test which evidence reaches its simulated answers. Deterministic practice simulation: every record and role is made up, and no AI model, external service or business system is used. Time: 30–45 minutes. Included with memberships that include the lessons.

Reliable events: prevent duplicate work and recover safely

Configure how a fictional event processor verifies, deduplicates, claims and retries work, then test ten fixed scenarios. Deterministic practice simulation: signature checks are fixture true/false values, the race is one in-memory interleaving and waiting time is virtual. No webhook, queue, payment or business system is connected. Time: 35–50 minutes. Included with memberships that include the lessons.

Evaluate output quality: catch misses without burying the team

Set a review threshold for a fictional reply router, then read the confusion matrix, metrics and category slices on twelve visible practice drafts. Risk scores and reference labels are hand-assigned for teaching; no AI model runs, nothing is sent and the drafts are not a held-out test set. Time: 30–45 minutes. Included with memberships that include the lessons.

More labs (planned, not built)