How to Build an AI Interview Question Generator

Generic interview questions get generic answers. I built an ai interview question generator that reads the job description, understands the role requirements, and produces targeted questions that actually reveal whether someone can do the work. Role-specific questions surface real skills.

This system generates behavioral, technical, and situational questions calibrated to what the role demands.

What You Need Before Starting

• Python 3.8+
• An AI API key (Claude)
• Job description text or structured role data
• An output format (markdown, JSON, or PDF)

Step 1: Define Question Categories

QUESTION_CATEGORIES = {
    "behavioral": {
        "description": "Past behavior predicts future performance",
        "count": 5,
        "format": "Tell me about a time when..."
    },
    "technical": {
        "description": "Role-specific knowledge and skills",
        "count": 5,
        "format": "Direct technical questions"
    },
    "situational": {
        "description": "Hypothetical scenarios the role will face",
        "count": 3,
        "format": "What would you do if..."
    },
    "culture": {
        "description": "Values and working style alignment",
        "count": 2,
        "format": "Open-ended preference questions"
    }
}

Step 2: Generate Questions from the Job Description

import anthropic
import json

client = anthropic.Anthropic()

def generate_interview_questions(job_description, categories):
    prompt = f"""Generate interview questions for this role. Each question should
reveal something specific about the candidate's ability to succeed.

Job Description:
{job_description}

Generate questions in these categories:
{json.dumps(categories, indent=2)}

For each question, include:
- The question itself
- What the question reveals (what good and bad answers look like)
- A follow-up probe question

Return as JSON with category keys."""

    message = client.messages.create(
        model="claude-sonnet-4-20250514",
        max_tokens=3000,
        messages=[{"role": "user", "content": prompt}]
    )
    return json.loads(message.content[0].text)

Step 3: Build a Scoring Rubric

def generate_rubric(questions_data):
    rubric = {}
    for category, questions in questions_data.items():
        rubric[category] = []
        for q in questions:
            rubric[category].append({
                "question": q["question"],
                "strong_answer_signals": q.get("good_answer", ""),
                "weak_answer_signals": q.get("bad_answer", ""),
                "score_range": "1-5",
                "weight": 2 if category == "technical" else 1
            })
    return rubric

Step 4: Format the Interview Guide

def format_interview_guide(questions_data, rubric):
    guide = "# Interview Guide\n\n"
    guide += "## Instructions\n"
    guide += "Score each answer 1-5. Take notes on specific examples given.\n\n"

    for category, questions in questions_data.items():
        guide += f"## {category.title()} Questions\n\n"
        for i, q in enumerate(questions, 1):
            guide += f"### {i}. {q['question']}\n"
            guide += f"**Look for:** {q.get('good_answer', 'N/A')}\n"
            guide += f"**Follow-up:** {q.get('follow_up', 'N/A')}\n"
            guide += f"**Score:** ___ / 5\n\n"

    return guide

Step 5: Adapt for Different Interview Rounds

def generate_round_specific(job_description, round_type):
    round_configs = {
        "phone_screen": "Brief questions to verify basic qualifications. 15 minutes.",
        "technical": "Deep technical assessment. Coding or problem-solving focus. 45 minutes.",
        "hiring_manager": "Role fit, team dynamics, growth potential. 30 minutes.",
        "culture": "Values alignment and working style. 20 minutes."
    }

    config = round_configs.get(round_type, "General interview. 30 minutes.")

    prompt = f"""Generate interview questions for a {round_type} round.
Context: {config}
Job: {job_description[:1000]}
Return 5 questions with scoring criteria as JSON."""

    message = client.messages.create(
        model="claude-sonnet-4-20250514",
        max_tokens=2048,
        messages=[{"role": "user", "content": prompt}]
    )
    return json.loads(message.content[0].text)

What to Build Next

Add candidate comparison. After all interviews, feed the scores into a ranking system that weights categories by what matters most for the role. The questions are the input. The hiring decision framework is the real system.

Related Reading

• The Automation Decision Tree - framework for deciding what to automate
• Input, Process, Output: The Universal AI Framework - structuring AI workflows
• Why AI Operations, Not AI Tools - building systems over collecting tools