I built an AI that "roasts" your resume (so recruiters won't)

Dobritoiu David-Constantin — Fri, 20 Feb 2026 11:30:51 +0000

Most "AI Resume Builders" are wrappers that just say: "Good job! Maybe add more numbers."
They give you a generic score (e.g., 72/100) but don't tell you why.
As a developer, I hate "Black Box" logic. If I get a low score, I want to see the stack trace. I want to see the diff.
So I built Refine.tools—a free, local-first tool that forces the AI to "Show its Math."
Here is how I engineered the "Audit Log" feature to give job seekers an unfair advantage.

The Problem: "Vague Feedback"

When you send a resume to an ATS (Applicant Tracking System), it parses your PDF into text and looks for specific keywords (e.g., "React", "TypeScript", "System Design").
If you miss them, you get rejected.
Most tools just use a generic GPT-4 prompt: "Review this resume."
The result is usually fluff.

The Solution: Gap Analysis with Structured JSON

I didn't want advice. I wanted a Gap Analysis.
I updated my backend (Next.js App Router) to accept two inputs:

The Resume (Parsed from PDF)
The Job Description (The target) Then, I used OpenAI's response_format: { type: "json_object" } to force a strict schema.

// The Structure I force the AI to return
interface ResumeAudit {
  missingKeywords: string[]; // "Critical skills found in JD but MISSING in Resume"
  formattingIssues: string[]; // "Header missing", "Photo detected"
  categories: {
    name: "Impact Language";
    score: number;
    feedback: string;
  }[];
}

By forcing this schema, the AI stops being a "Career Coach" and starts being a "Linter".
It returns arrays like:

"missingKeywords": ["AWS", "CI/CD", "Docker"]

The "Audit Log" UI

On the frontend, I render this as a strict "Pass/Fail" audit.

Red Badges: Skills you claimed to have but are missing from the text.
Green Checks: Formatting rules you passed (no columns, standard fonts). It turns the subjective art of "Resume Writing" into an objective "Bug Squashing" session.

Privacy First (The "Anti-SaaS")

The best part? I didn't want to store anyone's data.

Local-First: The resume parsing happens in your browser (or ephemerally in memory).
No Database: I don't confirm your email because I don't have a database.
No Sign-up: You just drag, drop, and refine.

Try it out

I built this because I believe the job market is broken, and candidates deserve better tooling than the recruiters rejecting them.
👉 Try the Resume Auditor (Free)
The strategy is simple: Don't guess. Audit.

I built a free AI resume tool with no database (because I hate signups)

Dobritoiu David-Constantin — Thu, 19 Feb 2026 18:19:56 +0000

I hate resume builders.

You know the drill. You spend 20 minutes typing in your work history, tweaking the formatting, and getting everything perfect. Then you hit "Download PDF" and get hit with a pop-up:

"Create an account to save."

Fine. You create an account.

"Upgrade to Premium for $19/month to download."

I got tired of this dark pattern. So I decided to build Refine.tools - a completely free, privacy-first alternative that runs entirely in your browser.

Here is how I built it without a database, and why "Local-First" is the future of these kinds of apps.

The "Anti-SaaS" Architecture

My goal was simple: Zero Data Liability.

If I don't store your data, I can't lose it, I can't sell it, and I don't need to ask for your email.

To do this, I had to ditch the standard "Postgres + Auth" setup.

1. The Database is Your Browser (Dexie.js)

Instead of saving your resume to my server, I save it to IndexedDB inside your browser.

I used a library called Dexie.js to make this easy. It acts just like a database table but lives 100% on your device.

This means you can close the tab, come back a week later, and your resume is still there. But I never see it.

2. The "Harvard" PDF Trick

Most resume tools use heavy JavaScript libraries to generate PDFs. This usually results in weird text selection issues or massive file sizes.

I took a different approach: Native CSS Print Styling.

I just used standard HTML and CSS for the resume layout (mimicking the classic "Harvard" template recruiters love). Then I used a simple @media print query to hide the buttons and navbars.

When you click "Download", the browser just prints the page to PDF. It is pixel-perfect, passes all ATS bots, and requires zero libraries.

Why this matters

We are drowning in "AI Wrappers" that are just data farms.

I wanted to prove you can build a useful, powerful tool that respects the user.

No Signup: You start using it instantly.
No Database: Your data is yours.
Free: I cover the small API costs because it's a passion project.

If you are currently job hunting, give it a try. It will score your resume against ATS filters and help you rewrite weak bullet points.

Check it out here: Refine.tools

DEV Community: Dobritoiu David-Constantin