Every founder learns this the expensive way: the correlation between a great resume and a great engineer is far weaker than it should be. You interviewed someone with three internships and a 9-point CGPA who couldn't debug a null pointer. You almost rejected someone with an empty-looking CV who turned out to be your best hire. The paper lied in both directions.
It lies because resumes are written to beat filters. Keywords get stuffed, group projects become solo triumphs, "worked on" stretches to cover things merely witnessed. Screening on resumes selects for resume-writing. If you want engineers, screen on engineering.
The 15-minute code review that replaces the resume screen
For each candidate, open their best repository and answer four questions:
- Did they build it? Read the commit history. Dozens of small commits with reasoning in the messages = a person thinking. One giant "final commit" = a person copying. This signal is nearly impossible to fake.
- Does it run? A live link or demo video that works is a stronger signal than any bullet point they could write about it.
- Can they explain it? A README that explains why — trade-offs, hard bugs, what they'd change — predicts how they'll communicate in your codebase.
- Is it real? A small tool with one genuine user outranks a big tutorial clone with none. Problem-finding is the skill you're actually hiring.
Fifteen minutes per candidate sounds expensive next to six seconds per resume — until you count the cost of interviewing the wrong ten people, or hiring the wrong one.
Make the interview structured, or it's just vibes
Unstructured interviews measure charm and interviewer mood. Decades of hiring research says structured ones — same questions, same rubric, scored independently — predict performance far better. Anchor the questions in the candidate's own work ("walk me through this design decision in your repo") and you get depth no LeetCode session reveals.
What this fixes beyond accuracy
Bias. College names and photos trigger shortcuts every human makes. Code reviewed on its own merits doesn't care where its author studied — which means you stop losing the brilliant tier-3 builder to a company that looked closer.
Your talent pool. The best proof-of-work candidates often have the worst resumes — they were building, not attending resume workshops. Screen on evidence and you're suddenly fishing where competitors aren't.
Or let the pipeline do it for you
Everything above is exactly what Iber productises: builders arrive with verified GitHub work, 60-second demos you can watch instead of resumes you have to decode, and structured AI interviews with integrity checks and comparable scores. You open a shortlist that's already ranked on proof.
The resume had a good fifty-year run. The evidence is better.