<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[ResumeForge Engineering]]></title><description><![CDATA[Engineering notes behind ResumeForge, an MIT open-source resume workspace built around strict evidence citations, non-AI checks against rounded or invented metr]]></description><link>https://resumeforge.hashnode.dev</link><generator>RSS for Node</generator><lastBuildDate>Sat, 21 Nov 2026 20:17:13 GMT</lastBuildDate><atom:link href="https://resumeforge.hashnode.dev/rss.xml" rel="self" type="application/rss+xml"/><language><![CDATA[en]]></language><ttl>60</ttl><item><title><![CDATA[Why AI Resume Tools Quietly Exaggerate Your Numbers (And How I Blocked It)]]></title><description><![CDATA[How I built a fast, non-AI code check in ResumeForge to block LLMs from rounding up numbers, inflating job titles, or inventing resume bullets.]]></description><link>https://resumeforge.hashnode.dev/why-ai-resume-tools-quietly-exaggerate-your-numbers-and-how-i-blocked-it</link><guid isPermaLink="true">https://resumeforge.hashnode.dev/why-ai-resume-tools-quietly-exaggerate-your-numbers-and-how-i-blocked-it</guid><category><![CDATA[AI]]></category><category><![CDATA[resume]]></category><category><![CDATA[internships]]></category><category><![CDATA[Open Source]]></category><category><![CDATA[TypeScript]]></category><dc:creator><![CDATA[Zenzer]]></dc:creator><pubDate>Sun, 15 Nov 2026 20:53:31 GMT</pubDate><enclosure url="https://cdn.hashnode.com/uploads/covers/6afa195f46f49b5a6ac40b21/8915114e-ebc1-40d8-8c48-373359d7a19f.gif" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>If you ask an AI to "tailor my resume to this job description without making anything up," it still tries to help you a little too much.</p>
<p>While building an open-source resume editor, I kept seeing models do things like:</p>
<ul>
<li>Rounding a <code>38%</code> speedup up to <code>40%</code></li>
<li>Dropping <code>Intern</code> from <code>Software Engineer Intern</code></li>
<li>Taking a real number (<code>10M requests</code>) and attaching it to something else (<code>10M users</code>)</li>
<li>Stretching an internship date range by an extra month</li>
</ul>
<p>None of these look like obvious lies when you skim the output. You usually only find out when an interviewer asks a follow-up question about a number you never actually hit.</p>
<p>After getting ghosted through fall recruiting as a second-year CS student, I built <a href="https://resumeforge.studio/"><strong>ResumeForge</strong></a> (<a href="https://github.com/ZenzerJs/ResumeForge">GitHub, MIT</a>). Here is how I stopped the AI from sneaking in exaggerated edits before anything can apply or export to PDF.</p>
<h2>Why "don't lie" in the prompt isn't enough</h2>
<p>Telling the AI to stick to the truth helps, and in ResumeForge every AI edit has to link back to an <strong>Evidence Bank</strong> of real bullets you added—or flag an honest <strong>Gap</strong> when you're missing a skill the job wants.</p>
<p>Still, prompts alone don't stop models from rounding <code>38%</code> to <code>40%</code>. And using a second AI call to check the first one adds wait time, costs extra tokens, and can miss the exact same small edits.</p>
<p>Instead, every AI suggestion runs through a plain code check against your original resume and Evidence Bank before you can apply or export it.</p>
<h2>What the code check actually looks for</h2>
<h3>1. Rounded numbers and swapped topics</h3>
<p>First, the checker pulls out every number, its unit (<code>%</code>, <code>ms</code>, <code>s</code>, <code>k</code>, <code>M</code>, <code>$</code>, <code>users</code>, etc.), and the word right after it from both your original resume and the AI's new version.</p>
<p>Then it checks that the exact number and unit actually exist in your real experience:</p>
<pre><code class="language-ts">const valueUnitMatches = masterFacts.metrics.filter((m) =&gt; {
  if (m.isTrivial) return false;
  const sameUnit = m.unit.toLowerCase() === candMetric.unit.toLowerCase();
  const sameValue = Math.abs(m.value - candMetric.value) &lt; 0.001;
  return sameUnit &amp;&amp; sameValue;
});

if (valueUnitMatches.length === 0) {
  violations.push({
    kind: "metric",
    severity: "HARD",
    field: "metrics",
    message: `Unverified or inflated metric claim "${candMetric.raw}" not supported by master evidence.`,
  });
}
</code></pre>
<p>If <code>38%</code> turns into <code>40%</code>, <code>valueUnitMatches</code> comes back empty and the edit is blocked. It also checks the noun after the number, so the AI can't take a real <code>10M requests</code> stat and turn it into <code>10M users</code>.</p>
<h3>2. Upgraded job titles</h3>
<p>Models love nudging your title closer to the job posting. To block that, the checker compares the words in your title using two simple rules:</p>
<ol>
<li>The AI can't add new title words that weren't in your original title (which blocks adding <code>Senior</code>, <code>Staff</code>, <code>Lead</code>, etc.).</li>
<li>If your real title includes words like <code>Intern</code>, <code>Co-op</code>, or <code>Junior</code>, the AI isn't allowed to drop them.</li>
</ol>
<p>So shortening <code>Full Stack Software Engineer</code> to <code>Software Engineer</code> is allowed, while turning <code>Software Engineer Intern</code> into <code>Software Engineer</code> gets blocked.</p>
<h3>3. Stretched dates</h3>
<p>Start and end dates (<code>May 2022 - Aug 2023</code>, <code>Jun 2024 – Present</code>) are compared against your original resume. The AI can never push a start date earlier, push an end date later, or change a finished job to <code>Present</code>.</p>
<h3>4. Honest gap flags</h3>
<p>When a job description asks for a tool you don't have in your resume or Evidence Bank, the AI is required to list it as an <strong>Explicit Gap</strong> instead of sneaking it into a bullet. If a new skill still shows up in an edit, the app flags it and asks you to confirm whether you actually know it.</p>
<h2>What still slips through right now</h2>
<p>Plain code rules are fast and free, but they still have a few blind spots I'm working on next:</p>
<ul>
<li><strong>Small plain numbers from 0 to 10 without units</strong> (like changing <code>team of 3</code> to <code>team of 5</code>—skipped right now so normal wording doesn't get falsely blocked, though <code>3x</code>, <code>5%</code>, and <code>12</code> <em>are</em> checked).</li>
<li><strong>Spelled-out numbers</strong> like <code>"forty percent"</code>.</li>
<li><strong>Adding a</strong> <code>+</code> <strong>sign</strong> (<code>40%</code> -&gt; <code>40%+</code>), since <code>+</code> currently gets stripped when reading the unit.</li>
</ul>
<h2>Try it out (no login or API key needed)</h2>
<p>ResumeForge is built with Next.js, TypeScript, PostgreSQL/Prisma, CodeMirror, and in-browser Typst WASM for live PDF previews. Outside of the pre-loaded demo mode, you can plug in your own API key (BYOK) without your key ever being saved on the server:</p>
<ul>
<li><strong>Live demo (no signup or API key):</strong> <a href="https://resumeforge.studio/">https://resumeforge.studio/</a></li>
<li><strong>Source code (MIT):</strong> <a href="https://github.com/ZenzerJs/ResumeForge">https://github.com/ZenzerJs/ResumeForge</a></li>
</ul>
<p>If you spot a clever way an AI could sneak a fake claim past these checks, let me know, I'd love to test it and patch the gap.</p>
]]></content:encoded></item></channel></rss>