Resume Matcher com Claude
nurbolsultanov/resume-matcher
Analisa fit currículo-vaga com IA e gera documentos otimizados
Ver no GitHubAplicação web que analisa compatibilidade entre currículo e descrição de vaga usando Claude AI e busca semântica. Gera pontuações de match, identifica palavras-chave faltantes, reescreve bullets e produz cartas de apresentação e PDFs otimizados para cada posição.
Instalação
npx convez-ai nurbolsultanov/resume-matcher1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Avaliar compatibilidade de currículo com vaga antes de candidatura
- Comparar múltiplas versões de currículo contra mesma JD
- Gerar cover letter e resume rewrite personalizados por posição
- Identificar gaps de skills e experiência antes de aplicar
Veredito

Ferramenta prática e bem-executada para otimizar candidaturas. Combina IA generativa (Claude) com embeddings semânticos de forma inteligente, entrega múltiplas perspectivas (global, por seção, comparativa) e gera artefatos úteis (PDF, cover letter). Stack maduro, sem dependências experimentais.
README
A tradução para o português deste README ainda está na fila.
AI Resume-JD Matcher
🚀 Live Demo: resume-matcher-nurbolsultanov.streamlit.app
A Streamlit web app that analyzes resume-JD fit using Claude AI and semantic search. Upload up to 3 resumes, paste a job description, and get actionable analysis across 4 tools.
What It Does
Tab 1 — Global Analysis
- Match Score (0-100) — how well your resume fits the JD
- Missing Keywords — top 5 terms in JD not found in your resume
- Bullet Rewrites — 3 suggested improvements tailored to the JD
- Gap Analysis — short summary of what's missing
Tab 2 — Section Analysis
- Individual scores for: Professional Summary, Technical Skills, Experience, Projects, Education
- Strengths, weaknesses, and one concrete suggestion per section
- Overall verdict
Tab 3 — Compare Resumes
- Upload 2-3 resume versions and compare them against the same JD
- Comparison table: match score, quantification score, missing keywords
- Per-resume details and final verdict: which resume to submit
Tab 4 — Generate Documents
- Cover Letter — tailored to the JD, downloads as .txt
- Resume Rewrite — section-by-section bullet rewrites with reasoning, what to add/remove, downloads as .txt
- Full Resume PDF — complete resume rewritten for this JD, auto-fits to 1 page, downloads as PDF
How It Works
┌──────────────┐ ┌──────────────┐ ┌──────────────────┐
│ Streamlit │ ──> │ pdfplumber │ ──> │ Resume bullets │
│ UI │ │ PDF parser │ │ (text chunks) │
└──────────────┘ └──────────────┘ └────────┬─────────┘
│
┌───────────────────────┴────────┐
│ │
▼ ▼
┌────────────────────┐ ┌─────────────────┐
│ sentence- │ │ JD keywords │
│ transformers │ │ extraction │
│ (MiniLM-L6-v2) │ └────────┬────────┘
│ embeddings + RAG │ │
└─────────┬──────────┘ │
│ │
└──────────────┬──────────────┘
▼
┌──────────────────────────┐
│ Anthropic Claude API │
│ (analysis + rewrites) │
└─────────────┬────────────┘
│
▼
┌──────────────────────────┐
│ Match score · keywords │
│ rewrites · gap analysis │
│ section scores · PDF │
└──────────────────────────┘
Tech Stack
- Python 3.11+
- Streamlit — UI
- Anthropic Claude — AI analysis and generation
- sentence-transformers — semantic retrieval (
all-MiniLM-L6-v2) - pdfplumber — PDF parsing
- reportlab — PDF generation
- pytest — testing
Setup
git clone https://github.com/nurbolsultanov/resume-matcher.git
cd resume-matcher
python -m venv venv
venv\Scripts\activate # Windows
source venv/bin/activate # Mac/Linux
pip install -r requirements.txt
Create .env:
ANTHROPIC_API_KEY=sk-ant-...
Run:
streamlit run app.py
Open http://localhost:8501.
Run Tests
pytest tests/ -v
Deploy on Streamlit Cloud
- Push repo to GitHub
- Go to share.streamlit.io
- Connect repo, set main file:
app.py - Add
ANTHROPIC_API_KEYin Settings → Secrets
Project Structure
resume-matcher/
├── app.py # Streamlit UI (4 tabs)
├── matcher.py # Claude API logic
├── pdf_parser.py # PDF text extraction
├── pdf_generator.py # Resume PDF generation (reportlab)
├── prompts.py # Claude prompts
├── requirements.txt
├── .gitignore
├── README.md
└── tests/
└── test_matcher.py
Limitations
- Match score reflects language overlap, not hiring probability
- Plain single-column PDFs parse best — heavy formatting loses ordering
- English only
- Each analysis uses
2-5K tokens ($0.01-0.05 per run)
About
Built by Nurbol Sultanov — Data Analyst in Los Angeles.
License
MIT