n8n Workflow: Validate Exam MCQs with AI Council (Gemini)

Generate and validate civil-services-style exam MCQs with Google Gemini—automatically

This n8n workflow manually generates a multiple-choice question and then validates it using Google Gemini. If the “Examiner” rejects the result, the workflow loops (up to three attempts) with reviewer feedback until you receive an approved JSON decision.

What this workflow does

• Manual trigger: Start by manually executing the workflow in n8n.
• Initialize run state: The Initialize State step sets the topic, an iteration counter (starting at 1), and optional feedback used to steer prompts.
• Generate → refine → evaluate: Google Gemini produces a draft MCQ, refines it, and then evaluates it as an “Examiner.”
• Parse Examiner JSON: The workflow extracts and parses the Examiner’s output, expecting a valid JSON object with at least a decision field (e.g., APPROVED/REJECTED).
• Iterate with feedback: If rejected and the iteration is under three, the workflow feeds the Examiner feedback into the next generation pass.
• Return final result: The workflow outputs the final approved content (or stops after the maximum attempts).

Use cases

• Create civil-services-style exam MCQs for practice sets while enforcing structured, reviewable output via JSON decisions.
• Validate question quality in an automation workflow for training content teams—reducing manual review cycles.
• Build an AI Council-style generation pipeline where an “Examiner” role tests outputs before accepting them.

Technical details

• Integration: Google Gemini (via n8nn8n-nodes-langchaingoogle gemini), used across Generator, Solver, and Examiner nodes.
• Workflow components: manual trigger, set, if, code, plus sticky note for clarity.
• Setup requirements: Add Google Gemini (PaLM) API credentials and choose your preferred Gemini model in each node.
• Input configuration: Provide at minimum a topic in the Initialize State node (with an iteration counter starting at 1, plus optional feedback).