👻 Ghost-Sourcer v6.0

100% Local AI Resume Screener for Technical Recruiters

---

🇧🇷 Português

Eu desenvolvi o Ghost-Sourcer para realizar triagens técnicas rodando 100% localmente. Como Recrutador Técnico, entendo que a privacidade dos dados dos candidatos é um pilar inegociável da nossa profissão. Esta ferramenta permite um processamento especializado diretamente no meu hardware, utilizando o poder do Llama 3 via Ollama.

🔒 Privacidade Total

• Zero dados na nuvem: Nenhuma informação sai da sua máquina
• Sem APIs externas: Todo processamento é local
• LGPD/GDPR nativo: Compliance por design

🚀 Por que eu criei esta ferramenta?

• Privacidade Total: O currículo nunca sai da sua máquina local.
• Equivalência Arquitetural: Vou além do mapeamento de palavras-chave; o agente analisa se as competências técnicas são transferíveis entre ecossistemas (ex: AWS → GCP).
• Custo Zero: Todo o processamento roda via Ollama na sua própria GPU.

🛡️ Por que IA Local (Edge)?

Em um ecossistema de recrutamento moderno, a soberania de dados é prioritária. O Ghost-Sourcer foi desenhado para atuar como uma camada de Edge Computing para triagem inicial.

• Ingestão Zero-Trust: Informações sensíveis (PII) permanecem em ambiente isolado.
• Sandbox Privado: Valide perfis sem trânsito de dados em redes externas.
• Mapeamento de Princípios: Foco em princípios de engenharia, não apenas keywords.

🛠️ Stack Técnica

• Runtime: Node.js v18+ com ES Modules
• Backend: Express.js
• AI Engine: Ollama + Llama 3 (100% local)
• Document Parsing: pdf-parse para PDFs

🖥️ Hardware Recomendado

• GPU: NVIDIA com 8GB+ VRAM (RTX 3060 ou superior)
• RAM: 16GB mínimo
• CPU: Qualquer processador moderno

---

🇺🇸 English

I built Ghost-Sourcer to run technical screenings 100% locally. As a Tech Recruiter, I recognize that candidate data privacy is a non-negotiable pillar of our field. This tool enables specialized processing by leveraging Llama 3 via Ollama directly on your own hardware.

🔒 Total Privacy

• Zero cloud data: No information leaves your machine
• No external APIs: All processing is local
• LGPD/GDPR native: Compliance by design

🚀 Why I Built This

• Total Privacy: Resumes never leave your machine.
• Architectural Equivalence: Beyond keyword matching - analyzes if skills transfer across ecosystems (e.g., AWS → GCP).
• Zero Cost: All processing runs via Ollama on your local GPU.

🛡️ Why Local Edge AI?

In modern recruiting, data sovereignty is priority. Ghost-Sourcer acts as an Edge Computing layer for initial screening.

• Zero-Trust Data Ingest: PII stays in an isolated local environment.
• Private Sandbox: Validate profiles without external network transit.
• Engineering-First Mapping: Focus on engineering principles, not just keywords.

🛠️ Tech Stack

• Runtime: Node.js v18+ with ES Modules
• Backend: Express.js
• AI Engine: Ollama + Llama 3 (100% local)
• Document Parsing: pdf-parse for PDFs

🖥️ Recommended Hardware

• GPU: NVIDIA with 8GB+ VRAM (RTX 3060 or better)
• RAM: 16GB minimum
• CPU: Any modern processor

---

📦 Installation

Quick Start (Windows)

1. Install Node.js

   • Download from nodejs.org (LTS version)
   • Run installer, click Next until done

2. Install Ollama

   • Download from ollama.ai
   • Run installer
   • Open terminal and run:

     ollama pull llama3

   • Wait for download (~4GB)

3. Setup Ghost-Sourcer

   git clone https://github.com/marcuscaiado/ghost-sourcer.git
   cd ghost-sourcer

   • Double-click SETUP.bat
   • Wait for dependencies to install

4. Run

   • Double-click START.bat
   • Browser opens automatically

Manual Installation (Mac/Linux)

# Clone repository
git clone https://github.com/marcuscaiado/ghost-sourcer.git
cd ghost-sourcer

# Install dependencies
npm install

# Start Ollama (in separate terminal)
ollama serve

# Pull model (first time only)
ollama pull llama3

# Start server
node server.js

# Open in browser
open http://localhost:3001

---

🚀 Usage

1. Paste Job Description - Include required level, location, skills, experience
2. Upload Resume PDF - Drag & drop or click to select
3. Click Analyze - Wait 10-30 seconds
4. Review Results - Score, verdict, gaps, recommendations

Scoring System

Score	Label	Verdict
90-100	UNICORN	ADVANCE
75-89	STRONG	ADVANCE
60-74	MAYBE	MAYBE
40-59	WEAK	REJECT
0-39	REJECT	REJECT

---

🔧 Configuration

Environment Variables

Variable	Description	Default
PORT	Server port	3001
OLLAMA_MODEL	Model to use	llama3

Using a Different Model

# Use Mistral instead
OLLAMA_MODEL=mistral node server.js

# Use Llama 3.1
ollama pull llama3.1
OLLAMA_MODEL=llama3.1 node server.js

---

❓ Troubleshooting

"Ollama is not running"

# Start Ollama service
ollama serve

"Connection Error" in browser

• Make sure you ran START.bat or node server.js
• Don't open index.html directly from file explorer

"Model not found"

# Pull the model
ollama pull llama3

Slow performance

• Ensure GPU drivers are updated
• Check that Ollama is using GPU: ollama ps
• Try a smaller model: OLLAMA_MODEL=llama3:8b

PDF not working

• Must be text-based PDF (not scanned image)
• Try re-exporting from original source
• Max file size: 15MB

---

🔐 Privacy Guarantee

Ghost-Sourcer is designed with privacy as the core principle:

• ✅ No cloud services - Everything runs locally
• ✅ No external APIs - No data leaves your machine
• ✅ No telemetry - No usage tracking
• ✅ No storage - Resumes are processed in memory only
• ✅ Open source - Audit the code yourself

Your candidates' data stays on YOUR machine. Period.

---

📄 License

MIT License - see LICENSE file.

---

👤 Author

Marcus Caiado

• LinkedIn: linkedin.com/in/marcuscaiado
• Website: marcuscaiado/resume/
• GitHub: github.com/marcuscaiado