Level 4 exam — No-Code & Automation

Exam — fresh questions

12 questions to validate the expert track: n8n and self-hosting, LLM calls, framed agents, documentary RAG, secret security and robustness.
Answer everything, then click Check: your score and explanations will appear.

4.1 · Recall

1. What is n8n's major distinctive feature compared to Make/Zapier?

It does no automation It charges per operation like the others It has no nodes Open source and self-hostable: total control over data, no per-operation billing

Sovereignty and controlled costs, at the price of a little more technicality.

4.1 · Application

2. Lina handles sensitive HR files with high volumes. What to recommend?

Prototype in the cloud then self-host with IT approval Stay on paper to be safe Send the files via personal email Self-host on an unmaintained PC, with no backups

Sensitive data + volumes = typical n8n case, self-host only if justified and framed.

4.2 · Recall

3. How do you integrate a model (ChatGPT, Claude, Gemini) into a workflow?

As a step: you send a prompt, you recover the answer for what follows By printing the prompt By changing the scenario's colour Impossible without coding

AI becomes a step between the trigger and the final action.

4.2 · Application

4. Lina classifies tickets: which instruction guarantees a reusable output?

Send the whole thread and attachments Answer in JSON with the category from the 3 imposed ones Answer in free verse Answer whatever you like, we'll see

Precise prompt + imposed JSON format = clean chaining to the router.

4.3 · Application

5. Lina's support agent must stay safe. Which setup?

Autonomous sending with no control No step limit and no trace Bounded goal, 3 tools (search, read, draft), the agent proposes and Lina validates Vague goal and all tools allowed

Limited tools, bounded goal, human in the loop for the sensitive.

4.4 · Recall

6. What is RAG for?

To colour answers To make AI answer from YOUR documents (search then generate) To delete data To print faster

Retrieve the relevant passages first, generate the answer second.

4.2 · Arbitration

7. The token cost of Lina's AI workflow explodes. What trade-off?

Send whole attachments on each call Call AI on every event with no filter Monitor nothing and increase volume Filter upstream, send only the useful text, concise answers, suitable model

Frugality (filter, size, concision) is designed when built.

4.3 · Arbitration

8. Should you roll out an AI agent for this need that chained calls already cover?

No: if well-chained AI calls suffice, the agent is superfluous Yes, without human validation to save time Yes, always, on principle Yes, with full rights to go fast

The agent is more powerful but less predictable: scale the tool to the need.

4.4 · Application

9. Lina builds her RAG on procedures. What is the right circuit?

Paste 300 pages into each prompt Answer from memory with no documents Question → retrieved passages → injected into the prompt → answer with sources, reindexing on updates Never update the base

Split, search, inject, cite — then reindex when docs change.

4.5 · Application

10. Where should Lina store her model's API key?

In plain text in the shared scenario In an email to the whole team In a demo screenshot In an encrypted connection / environment variable, never in plain text

A key is a password: safe place, minimal rights, rotation on doubt.

4.4 · Arbitration

11. Lina's RAG answers on payroll, a critical topic. What to decide?

Publish as is, RAG guarantees everything Keep systematic human control: RAG reduces inventions without removing them Remove sources to go faster Forbid any verification

Grounding is not guaranteeing: demanded sources + human on the critical.

4.5 · Arbitration

12. Lina's key leaked in a shared export on a Friday evening. First reflex?

Share a new copy with the key inside Wait until Monday to bother nobody Regenerate at once, limit rights, move to safety, check usage with IT Delete the export and hope

Regenerate first, analyse second — an exposed key is exploited within minutes.

What's next?

Score ≥ 80%: move to Level 5 — complete applications, industrialisation (versioning, documentation, governance) and ROI measurement.
Score 50-79%: review the failed sections (n8n 4.1, LLM calls 4.2, agents 4.3, RAG 4.4, security 4.5), re-read the explanations above, then retake the test.
Score < 50%: restart the level from 4.1 with the choice grid and the hardening checklist, then retry the exam with a fresh mind.