<?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: Aakash Das</title>
    <description>The latest articles on DEV Community by Aakash Das (@aakash_das).</description>
    <link>https://dev.to/aakash_das</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.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4036028%2Fcc547ab1-4dc9-4d04-af0f-61d2ed0e0b86.jpg</url>
      <title>DEV Community: Aakash Das</title>
      <link>https://dev.to/aakash_das</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/aakash_das"/>
    <language>en</language>
    <item>
      <title>Building an AI résumé scorer with Claude + AWS Bedrock - architecture and lessons learned</title>
      <dc:creator>Aakash Das</dc:creator>
      <pubDate>Sun, 19 Jul 2026 07:01:00 +0000</pubDate>
      <link>https://dev.to/aakash_das/building-an-ai-resume-scorer-with-claude-aws-bedrock-architecture-and-lessons-learned-4f7o</link>
      <guid>https://dev.to/aakash_das/building-an-ai-resume-scorer-with-claude-aws-bedrock-architecture-and-lessons-learned-4f7o</guid>
      <description>&lt;p&gt;Most résumés never reach a human recruiter. ATS software filters them out silently, with zero feedback to the applicant.&lt;/p&gt;

&lt;p&gt;That's the problem &lt;a href="https://www.scoraxis.com/" rel="noopener noreferrer"&gt;ScorAxis&lt;/a&gt; solves.&lt;/p&gt;

&lt;h2&gt;
  
  
  What it does
&lt;/h2&gt;

&lt;p&gt;Upload a PDF → get a 0–100 ATS compatibility score with specific issues flagged and actionable fixes.&lt;/p&gt;

&lt;h2&gt;
  
  
  Stack
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Backend:&lt;/strong&gt; FastAPI + AWS ECS Fargate + RDS Postgres&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Frontend:&lt;/strong&gt; Next.js 15 + Vercel + Clerk auth&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;AI:&lt;/strong&gt; Claude Haiku 4.5 (parsing) + Sonnet 4.6 (checks) via AWS Bedrock&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Jobs:&lt;/strong&gt; Inngest for background processing&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Scoring — two halves
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Deterministic (0–70)&lt;/strong&gt; — pure Python, no AI:&lt;br&gt;
  contact info, sections, word count, bullets, date consistency, layout density&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;LLM (0–30)&lt;/strong&gt; — Claude Sonnet via tool use:&lt;br&gt;
  action verbs, quantified impact, keyword coverage, tense, buzzwords&lt;/p&gt;

&lt;h2&gt;
  
  
  The interesting part — graceful degradation
&lt;/h2&gt;

&lt;p&gt;If Sonnet is unavailable, the upload never fails.&lt;br&gt;
The system persists a deterministic-only score with a "Limited Mode" banner. The user always gets something useful.&lt;/p&gt;

&lt;p&gt;Forced tool_choice made structured JSON output rock solid — almost zero validation failures in production.&lt;/p&gt;

&lt;h2&gt;
  
  
  Lessons
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Separate deterministic from AI logic from day one&lt;/li&gt;
&lt;li&gt;Forced tool use &amp;gt; prompt engineering for structured output&lt;/li&gt;
&lt;li&gt;Ship early — real feedback beats internal tuning&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;Live at &lt;a href="https://www.scoraxis.com/" rel="noopener noreferrer"&gt;scoraxis.com&lt;/a&gt; — happy to&lt;br&gt;
answer questions about the architecture below.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>architecture</category>
      <category>aws</category>
      <category>career</category>
    </item>
  </channel>
</rss>
