The AI Researcher Who Built Hiring Tools, Then Got Rejected By Them: Inside the Bizarre World Where AI Creators Can't Pass Their Own Systems

The Story That Should Terrify You (But Won't, After This)

You're applying to jobs. An algorithm screens your resume in milliseconds. You never hear back. Meanwhile, the person who built that algorithm? They just got rejected by it too.

This isn't fiction. This happened.

Dr. Prasad Rajpurohit, an AI researcher at a mid-sized recruiting tech firm, spent 18 months building a resume screening system. The tool used NLP, keyword matching, and behavioral scoring to identify "top-fit" candidates for enterprise clients. It was deployed across 400+ companies. It rejected thousands of applicants monthly.

Then Rajpurohit got laid off.

Needing work fast, he submitted his own resume to one of those 400 companies using the system he'd built. The system scored him at 34%. He didn't make the screening round. The hiring manager never saw his CV.

If you want to know exactly how these systems score your resume, here's the formula, reverse-engineered →

He knew exactly why. And that's the problem.

The Bizarre Physics of Algorithmic Rejection

Rajpurohit's failure wasn't about being underqualified. His resume contained the exact phrase architecture the tool was trained to recognize. He had every relevant keyword. His experience was directly applicable. But the system penalized him for something no human would care about: his résumé wasn't formatted in a way the parser expected.

His GPA section was missing (he'd graduated 15 years prior). The tool flagged this as "incomplete data." It downweighted his score.

His previous job title was "AI Research Lead" instead of the more conventional "Senior AI Engineer." The keyword matcher didn't recognize the mapping. Another penalty.

He'd listed his PhD but omitted the university name to save space. The degree validation module flagged it as suspicious.

None of these are real problems. All of them tanked his candidacy.

This isn't an edge case. This is the normal state of resume screening AI.

Hiring managers at those companies are frustrated too. They'll never know they rejected someone overqualified because a bot misread whitespace. The system is a black box. It's faster than humans. It's cheaper than humans. So it gets deployed. And qualified people disappear.

Why This Happens: The Training Data Problem Nobody Admits

Most hiring AI is trained on historical hiring data: resumes that got hired, resumes that didn't. The system learns patterns from what your company did before. But here's the trap—if your company has hiring biases (and every company does), the AI inherits them. Then it amplifies them.

If your historical hiring data favors certain universities, the algorithm will learn to overweight degrees from those schools. If it's biased toward specific job title formats, the system will penalize anything that deviates. If it's biased toward certain industries or resume structures, non-conforming candidates get filtered before humans ever see them.

Rajpurohit's tool was good relative to its training data. But its training data came from companies whose hiring managers had individual quirks, unconscious preferences, and structural biases. The AI didn't fix those. It weaponized them.

What Actually Works Against These Systems

Here's what Rajpurohit learned when he stopped being the victim and started being the hacker:

1. Keywords matter more than authenticity. If the job description uses "cloud infrastructure architect," that phrase needs to appear in your resume—even if you'd normally say "cloud platform engineer." ATS systems don't do semantic understanding well. They do pattern matching. Match the pattern.

2. Formatting is not decoration. Standard formatting (bullet points, clear section headers, consistent date formats) isn't about making your resume pretty. It's about making it parseable. Unusual fonts, graphics, tables, or unconventional layouts confuse the parser. You'll get penalized not for lack of skill, but for design choices.

3. Don't skip sections, even if they don't apply. A missing GPA section, a missing LinkedIn URL, a gap where a certifications line should be—these read as "incomplete data" to the system. Include the sections the algorithm expects, even if some are blank. (You can explain verbally in interviews.)

4. Title consistency is underrated. If a role is listed as "Senior Engineer" in the job description, try to use that exact title or a close variant in your resume. "Staff Engineer," "Principal Engineer," and "Senior Engineer" are semantically similar to humans. They're completely different to algorithms.

5. Quantify everything you can. Not for humans—algorithms love numbers too. Concrete metrics (30% improvement, 500K users, $2M revenue) signal specificity to the parser. Generic claims ("improved performance," "led team") are noise.

The Weird Truth: You're Not Competing Against AI. You're Competing Against the Training Data.

When Rajpurohit finally got through (a human recruiter manually reviewed his profile after a connection's referral), he asked the hiring manager what the system had scored him at. The manager pulled the report: 34%. The manager said, "That seems low for your background. Good thing someone else saw your application."

That manager was shocked when Rajpurohit explained the formatting issues. Shocked and frustrated. They'd been trusting the system. And the system had been filtering out decent candidates based on criteria nobody would consciously apply.

Here's the useful part of this story: you're not powerless. The system is dumb in specific ways. Once you understand how it's dumb, you can work around it.

Your resume shouldn't be built to impress a robot. It should be built to survive a robot so a human can actually read it. That's the whole game.

What to Do Monday Morning

Pull your resume right now. Check it against this list:

• Does it have standard section headers (Experience, Education, Skills)?
• Are dates formatted consistently (MM/YYYY or MM-YYYY, not "Spring 2019")?
• Do you use the exact keywords from the job description?
• Have you quantified your impact with numbers?
• Is there any unusual formatting, tables, or graphics that might break a parser?
• Did you skip any sections entirely? (Add them even if blank.)

If you're missing more than two of these, you're probably being filtered by algorithmic bias before humans ever see you.

Rajpurohit now advises job seekers. His first rule: "Don't assume the system is smarter than you. It's not. It's just faster and more rigid. Exploit the rigidity."

Your resume needs to pass through the robot. Then it needs to move a human. Build it to do both.