Français

🎓 Session 10 — Mock exam & final project

Applied AI · Advanced Level · Yann Isola — the capstone of the course

Preparation Claude Certified Architect

Exam simulator — 30 questions · 45 minutes

Actual conditions: stopwatch, no documentation, all questions answered (no penalty ⚠). The proposals are mixed up with each attempt . At the end: overall score, breakdown by domain and review mode with explanations. ⚠ Format aligned with the exam at the time of writing (60 q / 90 min / threshold 72%) — check the official Anthropic page.

Before you start

DomainWeight exam ⚠Questions
D1 · Architecture & agent orchestration27 %8
D2 · Tool design & MCP18 %5
D3 · Claude Code & workflows20 %6
D4 · Prompt engineering & structured output20 %6
D5 · Context & reliability15 %5

Two-pass method: answer everything that is certain, mark (⚑) the doubtful ones, come back to it. Pass threshold: 22/30 (≈ 72 %).

Certification Readiness Dashboard

Self-assess yourself on the 25 key skills (5 per area), cross-check with your mock exam score: the radar and the revision plan are built live. Be honest — the only loser from a flattering self-assessment is you on test day.

Self-assessment — 25 statements

For each statement: ✔ I know how to do (2 pts) · ~ To consolidate (1 pt) · ✘ To resume (0 pts)

Your readiness radar

Complete the self-assessment to generate the radar.

Prioritized revision plan

Priority = weakness × review weight ⚠. A weak domain at 27% comes before a weak domain at 15%.

Complete the self-assessment above.
Your decision (to write before leaving):
“I reserve the exam for the week of ___” Or “I review [areas] for ___ weeks, second blank on ___, then I reserve. »

The course map — 10 sessions

Click each session to review its key concepts, the remaining architect's reflex, and its connections with the rest of the course. This is your overall review sheet.

The 3 meta-lessons of the journey

1 · Deposit ≠ guaranteed. A prompt request; the code requires. Everything that is critical (permissions, validation, human approval) is applied programmatically.
2 · The least architectural effort. Call < chain < agent < multi-agent. We only go up one floor if there is proof of insufficiency of the lower floor.
3 · Measure before you believe. Evaluations, golden sets, provenance logs: intuition initiates, measurement decides.