Claude Projects

Projects I designed and built with Claude. Each one is filed with its source code and setup steps.

Brax10jns/Claude-Projects
No. 001 Resume Gauntlet ATS screen, then a four-reviewer panel, then a guided fix Node · Claude API · vanilla JS

Your resume, read by an ATS and four tough reviewers.

Resume Gauntlet screens a resume the way an applicant tracking system would, rewrites it to pass without inventing anything, then puts it in front of an interviewer, a rival applicant, a field expert and an executive. It ends with a resume you've fixed line by line.

Run it locally View source

It runs on your machine with your own Anthropic API key, so there's no hosted demo here. Setup takes about five minutes.

  1. 1
    ATS screenParse checks, required keywords verified against your text, measurable results and knockouts. Fails get rewritten and screened again, up to twice.
  2. 2
    Review panelFour reviewers, each with a temperament, weighted criteria and questions tied to specific lines.
  3. 3
    FixPick which flags to apply, fill in the details only you know, answer the reviewers' questions.
  4. 4
    Final resumeEdit by hand, find blind spots, download a PDF or text file, or run the gauntlet again.

One weak bullet, four readers

A line from the app's built-in sample resume, and the kind of pushback each reviewer gives it. The rewrite keeps the facts it has and leaves placeholders for the ones only you know.

Sample resume · Data Analyst, 2021 to present

• Responsible for creating reports for the operations team

After the fix step

Built operations reports and Excel/Tableau dashboards used by [ADD: number] stakeholders, cutting manual prep by [ADD: hours per week].

  • The Interviewer

    Which reports, for whom, how often? This is the first thing I'd ask you about.

  • The Rival Applicant

    Every analyst in the pile writes this line. Mine has a number in it.

  • The Field Expert

    Name the tools and the data behind them. "Reports" could be a pivot table or a pipeline.

  • The Executive

    What changed for the business because these reports existed?

Illustrative example written for this page, not output from a real run.

The panel

Each reviewer reads the version that passed the ATS screen, scores it on their own weighted criteria, and flags issues by quoting your actual lines. A flag that quotes text not in your resume is thrown out.

  • The Interviewer

    Hiring manager deciding whether to interview you

    Skeptical but fair. Looks for what can be verified and where the story doesn't add up, and asks the questions you'd face in the room.

    • Credibility of claims ×1.5
    • Worth an interview slot ×1.5
    • Consistency, story, results ×1

  • The Rival Applicant

    A strong candidate applying for the same job

    Competitive and blunt. Compares you with the top 10% of applicants and names where they'd beat you.

    • Stands out from the stack ×1.5
    • Proof of results ×1.5
    • Tools, credentials, memorability ×1

  • The Field Expert

    Senior practitioner in your field

    Precise and a bit pedantic. Judges whether you know the work at the level you claim, and catches outdated tools and misused jargon.

    • Technical depth shown ×1.5
    • Core competencies covered ×1.5
    • Tools, terminology, scope ×1

  • The Executive

    VP who signs off on the hire

    Impatient and commercially minded. Skims the top third in 30 seconds and asks what you're worth to the business.

    • Business impact ×2
    • Scope and ownership ×1.5
    • Case for the salary ×1.5

The score is math, not a vibe

Scores are computed in the page, not by the model, so every attempt is judged the same way.

  • Pass mark 75 with no critical flags. A failing resume is rewritten truthfully and screened again, up to two rewrites.
  • Keywords are verified. A keyword only counts when its words actually appear in your resume.
  • Reviewer scores are each reviewer's weighted criteria, rated 1 to 10 and scaled to 100.
  • Checks that don't need AI run on every screen: missing contact details, weak openers, buzzwords, first person, mixed date formats, length.
  • Nothing is invented. Rewrites never add employers, dates, tools or numbers. Missing facts become placeholders like [ADD: % time saved].
  • Reads your files in the browser: PDF, Word .docx (including headers, tables and text boxes) and plain text. Exports PDF or text.

Run it locally

You need Node.js 18 or later and an Anthropic API key. The key stays on your machine and is never sent to the browser.

  1. Get the code

    git clone https://github.com/Brax10jns/Claude-Projects.git
cd Claude-Projects/resume-gauntlet
npm install

  2. Save your API key

    On Windows, this asks for the key without showing it and stores it in your user profile, outside the project folder. Then open a new terminal.

    npm run set-key

    On macOS or Linux, set it in your shell, or copy .env.example to .env and fill it in.

    export ANTHROPIC_API_KEY=your-key-here

  3. Start it

    npm start

    Then open http://127.0.0.1:3000 and upload a resume or paste one in.

  4. Optional settings

    VariableDefaultWhat it does
    CLAUDE_MODELclaude-opus-5-5The Claude model used for every review.
    PORT3000Port the server listens on.
    HOST127.0.0.1Interface to listen on. Keep it local unless it sits behind your own login.

Cost

  • Each round makes about 6 to 11 Claude calls: screens, rewrites, four reviews and a final rewrite.
  • Billed to your own Anthropic account at standard API rates.

Privacy

  • Your resume goes to Anthropic's API only when a review runs. The server keeps no copies and logs no content.
  • Results and drafts stay in your browser's local storage.

Limits

  • The ATS screen simulates how tracking systems and recruiters typically filter. Real systems differ by employer.
  • The reviewers are AI personas and can be wrong. Check every change before you send your resume anywhere.