🚀 LinkedIn Sourcing Agent with Candidate Scoring

A comprehensive AI agent that sources LinkedIn profiles, scores candidates using a sophisticated fit score algorithm, and generates personalized outreach messages.

🎯 Features

✅ Step 1: LinkedIn Profile Discovery

• PDF Job Description Processing: Extracts text from PDF job descriptions using OCR
• Intelligent Search: Uses Google search to find relevant LinkedIn profiles
• Robust Parsing: Handles various LinkedIn URL formats and profile structures
• Caching System: Implements SQLite caching to avoid re-fetching data

✅ Step 2: Candidate Scoring Algorithm

• Comprehensive Fit Score: 1-10 rating based on 6 key criteria
• Weighted Scoring System:
  • Education (20%): Elite schools, technical degrees
  • Career Trajectory (20%): Progression, seniority levels
  • Company Relevance (15%): Top tech companies, industry match
  • Skills Match (25%): Technical skills alignment
  • Location Match (10%): Geographic compatibility
  • Tenure (10%): Job stability and progression

✅ Step 3: Message Generation

• Personalized Outreach: AI-generated messages referencing candidate details
• Professional Tone: Maintains recruiter-friendly communication style
• Customization: Tailored to specific job requirements and candidate strengths
• Message Analysis: Effectiveness scoring and optimization
• Multiple Variations: A/B testing capabilities

🏗️ Architecture

Job Description PDF → Text Extraction → LinkedIn Search → Profile Scoring → Message Generation → Results
       ↓                    ↓                ↓              ↓              ↓              ↓
   OCR/PDF2Image    Google Search API   BeautifulSoup   Fit Algorithm   AI Templates   JSON Output

📊 Scoring Rubric

Education (20%)

• Elite Schools (MIT, Stanford, Harvard, etc.): 9-10
• Strong Technical Schools: 7-8
• Standard Universities: 5-6
• Relevant Degrees (CS, Engineering, AI/ML): +1 bonus

Career Trajectory (20%)

• Director/Principal Level: 9.0
• Lead/Senior Level: 7.5
• Mid-Level: 6.5
• Entry Level: 5.0

Company Relevance (15%)

• Top Tech Companies (Google, Microsoft, etc.): 9.0
• AI/ML Companies: 8.0
• Tech Startups: 7.0
• Other Companies: 6.0

Skills Match (25%)

• 5+ Matching Skills: 9.5
• 3-4 Matching Skills: 8.0
• 1-2 Matching Skills: 6.5
• No Matches: 5.0

Location Match (10%)

• Exact City Match: 10.0
• Metro Area Match: 8.0
• Remote-Friendly: 6.0
• No Match: 5.0

Tenure (10%)

• 1-2 Companies (Stable): 9.0
• 3-4 Companies (Reasonable): 7.0
• 5-6 Companies (Some hopping): 5.0
• 7+ Companies (Frequent changes): 3.0

🚀 Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Run the Complete Pipeline

python main_integrated.py

3. Test Message Generation

python test_messages.py

4. Test Complete Pipeline

python main_integrated.py

5. Run Individual Components

# LinkedIn search only
python run.py

# Scoring with sample data
python test_scoring.py

# Message generation only
python test_messages.py

📁 Project Structure

AI_challenge/
├── main_integrated.py          # Complete pipeline
├── run.py                      # LinkedIn profile finder
├── candidate_scorer.py         # Scoring algorithm
├── message_generator.py        # Message generation
├── test_scoring.py            # Test script with sample data
├── test_messages.py           # Test script for messages
├── linkedin_agent.py          # Basic LinkedIn search
├── main.py                    # Simple entry point
├── requirements.txt           # Dependencies
├── README.md                  # This file
├── sourcing_results.json      # Output results
├── linkedin_cache.db          # SQLite cache
└── *.pdf                      # Job description PDFs

🔧 Configuration

Elite Schools List

The system recognizes top-tier institutions for education scoring:

• MIT, Stanford, Harvard, Berkeley, Caltech, CMU, Princeton
• Yale, Columbia, UPenn, Cornell, Brown, Dartmouth, Duke
• Northwestern, UChicago, NYU, USC, UCLA, UCSD, Georgia Tech
• UIUC, UMich, UW, UT Austin, and more

Top Tech Companies

Recognized companies for company relevance scoring:

• Google, Microsoft, Apple, Amazon, Meta, Netflix
• Uber, Lyft, Airbnb, Stripe, Palantir, Databricks
• OpenAI, Anthropic, DeepMind, Waymo, Tesla, SpaceX
• And 30+ other top tech companies

AI/ML Skills

Comprehensive skill matching for technical roles:

• Programming: Python, C++, Java, Julia, R, MATLAB
• ML Frameworks: TensorFlow, PyTorch, Scikit-learn, Keras
• AI/ML: NLP, Computer Vision, Deep Learning, LLMs
• Tools: AWS, Azure, GCP, Docker, Kubernetes, Spark
• And 40+ other relevant skills

📊 Sample Output

{
  "job_id": "ai-ml-research-windsurf",
  "candidates_found": 8,
  "top_candidates": [
    {
      "name": "Dr. Sarah Chen",
      "linkedin_url": "https://linkedin.com/in/sarah-chen-ml",
      "fit_score": 8.7,
      "score_breakdown": {
        "education": 9.5,
        "trajectory": 8.0,
        "company": 9.0,
        "skills": 9.5,
        "location": 8.0,
        "tenure": 7.0
      },
      "headline": "Senior ML Engineer at Google | PhD Stanford"
    }
  ],
  "outreach_messages": [
    {
      "candidate_name": "Dr. Sarah Chen",
      "linkedin_url": "https://linkedin.com/in/sarah-chen-ml",
      "fit_score": 8.7,
      "message": "Hi Sarah, I came across your impressive background in Python and your work at Google...",
      "key_highlights": {
        "education": "Stanford University",
        "top_company": "Google",
        "key_skill": "Python"
      }
    }
  ]
}

🛠️ API Integration

The system provides API-ready response formats:

from main_integrated import LinkedInSourcingAgent

agent = LinkedInSourcingAgent()
results = agent.process_job_description("job_description.pdf")
api_response = agent.get_api_response_format(results)

🌐 API Deployment

Local Development

# Start the FastAPI server
python api.py

# Test the API
python test_api.py

Hugging Face Spaces Deployment

1. Create Space: Go to huggingface.co/spaces
2. Upload Code: Push your code to the Space repository
3. Access API: Your API will be available at https://your-space-url.hf.space/api

API Endpoints

• Health Check: GET /api/health
• Process Job: POST /api/process-job
• Process PDF: POST /api/process-pdf
• Batch Process: POST /api/batch-process
• Documentation: GET /api/docs

Interactive Demo

Visit your Hugging Face Space URL for an interactive Gradio interface.

For detailed deployment instructions, see deployment_guide.md.

🔄 Caching System

• SQLite Database: Stores search results for 24 hours
• Automatic Cache Management: Prevents duplicate requests
• Configurable TTL: Adjustable cache expiration

🚨 Rate Limiting

• Intelligent Delays: Random delays between requests (1-5 seconds)
• User-Agent Rotation: Realistic browser headers
• Error Handling: Graceful handling of blocked requests

🔮 Future Enhancements

Planned Features

• Message Generation: AI-powered personalized outreach
• FastAPI Endpoint: Deployable API service
• Batch Processing: Handle multiple jobs simultaneously
• Multi-Source Integration: GitHub, Twitter, personal websites
• Confidence Scoring: Uncertainty quantification
• Advanced Analytics: Detailed candidate insights

Technical Improvements

• LinkedIn API Integration: Official API access
• Enhanced OCR: Better PDF text extraction
• Machine Learning: Learn from scoring patterns
• Distributed Processing: Scale to 100s of jobs

🤝 Contributing

This project is part of the Synapse AI Hackathon Challenge.

Development Guidelines

1. Use Cursor: All development should be done in Cursor IDE
2. Python Focus: Primary language is Python
3. Documentation: Comment code and explain decisions
4. Testing: Include test cases for new features
5. Performance: Consider scalability and rate limiting

📝 License

This project is created for educational purposes as part of the Synapse AI Hackathon Challenge.

🆘 Troubleshooting

Common Issues

PDF Text Extraction Fails

• Ensure PDF is not password protected
• Try different PDF files
• Check if PDF contains scanned images

No LinkedIn Profiles Found

• Try broader search terms
• Check internet connection
• Verify Google search is accessible

Scoring Errors

• Ensure all dependencies are installed
• Check candidate data format
• Verify job description content

Debug Mode

Enable debug output by modifying the search functions to print detailed information about the search process.

📞 Support

For questions about this implementation:

• Check the code comments for detailed explanations
• Review the test scripts for usage examples
• Examine the scoring algorithm for customization options

---

Built with ❤️ for the Synapse AI Hackathon Challenge