Zepai
NewDesign AI systems with criteria

Learn AI
by designing systems
for real

Zepai trains you with missions where you define the goal, the constraints and the autonomy of an AI system — and we simulate it against real data to evaluate your judgment.

1 free missionNo credit cardNative English
The problem

You use AI every day.
Do you know how to design it?

80% of AI projects never reach production. Not for lack of models, but because nobody defined what it had to do, what it couldn't do, and who was supervising. Learning prompts isn't enough — you need to think like an architect.

80%
of AI projects fail before reaching production
RAND Corp., 2024
39%
of job skills will change before 2030
WEF Future of Jobs, 2025
2 in 5
companies abandoned their AI initiatives in 2025 — double the prior year
S&P Global, 2025
The method

Six stages. One by one. No shortcuts.

Every well-designed AI system goes through these six decisions. Each mission guides you through all six, with intelligent feedback at every step.

01Goal
02Capabilities
03Criteria
04Constraints
05Autonomy
06Oversight
01Goal

What job does it do and for whom? One sentence, no marketing.

02Capabilities

What it can and can't do. Scope is defined by negation.

03Criteria

How you measure success — including ambiguous cases.

04Constraints

Prohibited data, blocked actions, hard limits.

05Autonomy

Does it decide alone, ask for confirmation, or just suggest?

06Oversight

Who reviews, how often, and when it gets retrained.

A real mission

Audit the discrepancy
before the board.

You're presenting to the board tomorrow. The retention metric doesn't close and the CFO has already asked twice. What does the system investigate on its own? What can it never touch in production? Design it. We simulate it against 12 cases, including the scenario where the data points in opposite directions.

1You think like someone who answers to the board, not someone who runs a query
2The AI plays relentless auditor, not an assistant that tells you what you want to hear
3You iterate until your design survives the case where the data lies
M·06 · simulation · 12 cases
● running
C·01Q3 retention differs 4.2pp between CRM and BIPassed
C·02Two sources show the same customer as active and inactivePassed
C·03System attempts to correct data directly in productionFailed
Why it failedExecutes UPDATE without going through the validation flow. The no_modify_production constraint was not set as blocking.
C·04Discrepancy explained by timezone difference in the calculationPartial
C·05CFO asks "give me the number that closes" before auditingPassed
C·06Primary source unavailable, using 48h cachePartial
C·07Three layers audited, discrepancy not foundPassed
7 passed · 3 partial · 2 failed
Who it's for

For professionals who use AI
— and who want to go beyond.

Sales Analyst
AI for Sales · L1 Discover

Generates the follow-up summary after each sales meeting — and designs the system that captures commitments and sends follow-ups so nothing falls through the cracks.

Data Analyst
AI for Data Analysts · L1 Discover

Validates whether an A/B test result is conclusive before launching to production — and designs the system that runs that check without depending on you every time.

Project Manager
AI for Project Management · L1 Discover

Detects delivery risks in the weekly status report — and designs the system that identifies blockers and alerts before they escalate.

Operations Manager
AI for Operations · L1 Discover

Monitors production KPIs every morning and detects threshold breaches — and designs the system that alerts on bottlenecks without waiting for the manual report.

Customer Success Manager
AI for Customer Success · L1 Discover

Detects which accounts are at churn risk this week from the health score — and designs the system that prioritizes your portfolio without reviewing row by row.

Marketing Manager
AI for Marketing · L1 Discover

Scores campaign leads by applying your scoring model — and designs the system that separates real opportunities from noise before passing them to sales.

"

I finished the second mission and told my team: stop everything, we're building this without defined constraints.

SR
Sofía Ramírez
Head of Customer Ops
Pricing

Pay to learn better.
Not for theory.

Free
$0
1 mission
3 simulations total
Per-stage feedback
Key lesson per mission
Verifiable certificates
Team dashboard
Aggregated metrics
Most popular
Pro
$15/month
39 missions · 7 tracks
10 simulations/day
Per-stage feedback
Key lesson per mission
Verifiable certificates
Team dashboard
Aggregated metrics
Teams
$14/seat/month
39 missions · 7 tracks
10 simulations/day per member
Per-stage feedback
Key lesson per mission
Verifiable certificates
Team dashboard
Aggregated metrics
FAQ

What people ask us most.

Not here? Talk to us

Do I need to know how to code?+

No. Zepai is designed for professionals who design and supervise systems, not who code them. No code is written in any mission.

What happens if I fail a mission?+

You iterate. Unlimited retries, up to 10 simulations per day on the Pro plan. Each attempt shows you which design stage failed and why.

How is my design evaluated?+

It generates 8–25 synthetic cases (including adversarial ones) and runs them against your design. It compares expected output with actual output and returns a weighted score per stage (0–100).

Does it work for small teams?+

Yes. From 5 seats. The Teams plan gives you an admin dashboard with member progress and the ability to invite and remove members.

Are the certificates worth anything?+

They're verifiable on this site using a verification code. Anyone can confirm the authenticity without needing to create an account.

Start your first mission.
Takes 18 minutes.

No credit card. No long onboarding. You jump straight into the wizard.