<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Seong Il Gang</title>
    <description>The latest articles on DEV Community by Seong Il Gang (@ait426).</description>
    <link>https://dev.to/ait426</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3858840%2F1fe373f4-2619-42cc-8575-b31392d58d03.png</url>
      <title>DEV Community: Seong Il Gang</title>
      <link>https://dev.to/ait426</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/ait426"/>
    <language>en</language>
    <item>
      <title>I built a self-evolving AI skill engine as a solo dev using 4 AI models</title>
      <dc:creator>Seong Il Gang</dc:creator>
      <pubDate>Fri, 03 Apr 2026 06:11:17 +0000</pubDate>
      <link>https://dev.to/ait426/i-built-a-self-evolving-ai-skill-engine-as-a-solo-dev-using-4-ai-models-23c8</link>
      <guid>https://dev.to/ait426/i-built-a-self-evolving-ai-skill-engine-as-a-solo-dev-using-4-ai-models-23c8</guid>
      <description>&lt;h2&gt;
  
  
  What is Cambrian?
&lt;/h2&gt;

&lt;p&gt;An AI skill engine where skills get better the more you use them.&lt;/p&gt;

&lt;p&gt;Every AI tool runs the same prompts forever. Cambrian is different:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Give feedback → LLM mutates the skill&lt;/li&gt;
&lt;li&gt;3-trial benchmark → LLM Judge blind-scores A/B&lt;/li&gt;
&lt;li&gt;Better version survives, worse one dies&lt;/li&gt;
&lt;li&gt;After 10 rounds, the skill is personalized to YOUR workflow&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  How I built it
&lt;/h2&gt;

&lt;p&gt;Solo Korean developer. 4 AI models in relay:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Claude Opus&lt;/strong&gt; → architecture &amp;amp; critical thinking (40 rounds of self-critique)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;GPT&lt;/strong&gt; → code generation&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Claude Sonnet&lt;/strong&gt; → implementation &amp;amp; debugging&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Gemini&lt;/strong&gt; → review&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The project itself was evolved through its own philosophy:&lt;br&gt;
feedback → improve → test → repeat.&lt;/p&gt;

&lt;h2&gt;
  
  
  Key features
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;3-channel feedback&lt;/strong&gt;: auto (failure analysis), critic (proactive review), manual&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;LLM Judge&lt;/strong&gt;: blind A/B scoring (0-10) with reasoning&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Competitive execution&lt;/strong&gt;: same domain, best skill wins&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Multi-provider&lt;/strong&gt;: Claude, GPT, Gemini — not locked to one platform&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Skill decay&lt;/strong&gt;: 30 days unused → dormant, 90 days → fossil&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Export/import&lt;/strong&gt;: portable .cambrian packages&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Numbers
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;180 tests, 0 failures&lt;/li&gt;
&lt;li&gt;14 seed skills&lt;/li&gt;
&lt;li&gt;16 CLI commands&lt;/li&gt;
&lt;li&gt;3 LLM providers&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Try it
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;pip &lt;span class="nb"&gt;install &lt;/span&gt;git+https://github.com/Ait426/cambrian.git
cambrian init &lt;span class="nt"&gt;--dir&lt;/span&gt; ./my_project
cambrian skills
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;GitHub: &lt;a href="https://github.com/Ait426/cambrian" rel="noopener noreferrer"&gt;https://github.com/Ait426/cambrian&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;What would you break first?&lt;/p&gt;

</description>
      <category>ai</category>
      <category>python</category>
      <category>opensource</category>
      <category>productivity</category>
    </item>
  </channel>
</rss>
