AI interview hacks · Lesson 4 of 6 · 15 min
Reverse the job post, then cross-examine your resume
Part 1Basics
A job post is a list of what the team worries about. Count what it mentions and you get a rough map of where the questions will come from.
Your resume is the other source. Anything on it is fair game, and follow-ups go deep. An interviewer at Kapwing described a candidate with a polished, keyword-heavy resume who was fluent at first, then couldn't explain why a load spike happened, or what data a claimed feature handled.
Here's the rule that matters: use AI to pressure-test your real experience, and never to invent experience. If you can't defend a line through five follow-ups, fix your understanding or remove the line.
Words you need
- Demand
- How often a skill shows up in the post. A rough signal, not a fact.
- Priority
- Demand multiplied by how weak you are at it. That's what you study first.
- Follow-up ladder
- Why, how, what broke, what would you change, what would it cost? Interviewers walk down it.
A worked example
Meera's job post mentions retrieval four times, evals three times and agents once. She'd planned to spend the week on agents. The numbers move RAG to the top. Then she pastes one resume line about a search feature into a cross-examination prompt, and by the third question she can't say what recall it reached. She goes and finds out.
Part 2The question
How do you find out what they'll ask, and whether your own stories hold up?
By the end of this lesson, you should be able to:
- Turn a job post into a study order
- Survive five follow-ups on any resume line
- Spot resume lines to fix or remove
Part 3In detail
Step by step
- Paste the post into the mapper. Or ask AI to list the skills, with quotes from the post.
- Score yourself from 1 to 5. Be honest. A later mock will correct you anyway.
- Study by priority. Demand times weakness tells you where a day pays off most.
- Cross-examine each resume line. One line at a time, up to five follow-ups.
- Fix the gaps with facts. Look up the real numbers from your work. Don't let AI supply any.
Don't let AI write your story
A model will happily produce a stronger-sounding story, with invented numbers. That's how a candidate ends up silent at the follow-up. Ask it for questions, and keep the answers yours.
Prompts to copy
Cross-examine one line
Run this for every line you'd be nervous about.
You are a skeptical senior engineer interviewing me. Here is one line from my resume: [line] Ask one follow-up at a time. Go deeper with each: why that approach, how it worked, what broke, what you measured, what you'd change. Don't help me or suggest answers. After 5 questions, list exactly where my answers got vague.
Read the job post like the hiring manager
Ask for quotes, so you can check that it didn't make things up.
Here is a job post: [post] List the skills it really cares about, ranked. For each, quote the words that show it and write the two interview questions most likely to test it. Say which skills are only implied. Don't guess the company's process.
The code, explained
This is real code, and our tests run it, including the buggy versions, to prove the bugs are real.
Counting what the post asks for
A word count with care taken over word edges, so 'rag' doesn't match 'fragrance' or 'storage'.
1export function mapJobPost(post: string, rules: SkillRule[]): [string, number][] {2 const out: [string, number][] = [];3 for (const { skill, keywords } of rules) {4 const hits = keywords.reduce((n, k) => n + (post.match(new RegExp(`(^|[^a-z0-9])${escape(k)}(?![a-z0-9])`, "gi")) ?? []).length, 0);5 if (hits > 0) out.push([skill, hits]);6 }7 return out.sort((a, b) => b[1] - a[1] || a[0].localeCompare(b[0]));8}(^|[^a-z0-9])- The match must start at a word edge.
(?![a-z0-9])- And must end at one.
escape(k)- Keeps a keyword such as 'c++' from being read as a regular expression.
Tested by npm run test:modules.
From demand to a study order
Combines what they want with what you lack.
1/** Priority is demand times weakness, where 5 means strong. A skill you've nailed scores 0 and drops off. */2export function studyOrder(demand: [string, number][], myScores: Record<string, number>): { skill: string; priority: number }[] {3 return demand4 .map(([skill, hits]) => ({ skill, priority: hits * (5 - (myScores[skill] ?? 1)) }))5 .filter((x) => x.priority > 0)6 .sort((a, b) => b.priority - a.priority || a.skill.localeCompare(b.skill));7}hits * (5 - (myScores[skill] ?? 1))- A skill you haven't scored counts as weak, since not knowing is a reason to look.
.filter((x) => x.priority > 0)- A skill at 5 of 5 drops off the list, however often it's mentioned.
Tested by npm run test:modules.
What can go wrong
- Treating the map as the truth
- It's a count of words. Ask your recruiter what the rounds are, too.
- Keeping a line you can't defend
- A confident line followed by silence does more damage than leaving the line out.
Part 4Questions and answers
Answer out loud first, then open each one.
Why multiply demand by weakness?
What should you never ask AI to do to your resume?
Part 5Going further
Drill: One line, five follow-ups (15 min)
- Pick the resume line you're least sure about.
- Run the cross-examination prompt, answering out loud.
- Write down where you got vague.
- Find the real facts and answer again.
Keep the fundamentals fresh with the critical basics.
Who says so
Checked in November 2026. Vendors and individuals often report what candidates told them, so treat their details as reports.
- What it's like to interview a software engineer preparing with AI Individual, April 2025
Next: the habit that ties the other hacks together. Every miss gets written down.