Apppublic

IsmailGhediri/synapse-recruiter-api

sourceHugging Faceupdated 1y agoView on Hugging Face

0likes

App README

Synapse Recruitment Automation

This work takes part of Synapse Annual First Ever AI Hackathon - Sourcing Agent Challenge.

It's an automated pipeline that:

1.📝 Processes job descriptions
2.🔍 Finds matching LinkedIn candidates, caches them to avoid refetching
3.💯 Scores candidates intelligently
4.✉️ Generates personalized outreach

Advanced Cache Features

python

# cache.pkl structure (automatically maintained)
{
    "linkedin_url": {
        "profile": {candidate_data},
        "timestamp": "2023-11-20T14:30:00"  # 7-day expiration
    }
}

🚀 Quick Start

1. Clone repo

bash

git clone https://github.com/ismailghediri/synapse-sourcing-agent.git
cd synapse-ai-sourcing-agent

2. Install dependencies

bash

pip install -r requirements-api.txt

3. Configure API Keys (Optional for Local Use)

Create a .env file in the root directory with your own API keys:

bash

SERPER_API_KEY=your_serper_api_key
GEMINI_API_KEY=your_gemini_api_key

Note: .env is ignored in .gitignore to prevent exposure. Not required if using the Space.

4. Run Locally

bash

uvicorn app:app --reload

Access the FastAPI UI at http://127.0.0.1:8000/docs to test endpoints. Without keys, test mode activates.

5. Run with Docker:

bash

docker build -t synapse-agent .
docker run -p 8000:8000 synapse-agent

Add your .env file to the container if using local keys.

🛠️ Usage

Hosted on Hugging Face Spaces with an Interactive UI: https://ismailghediri-synapse-recruiter-api.hf.space/docs Uses Space Secrets for SERPERAPIKEY and GEMINIAPIKEY, enabling full functionality without local configuration.

Or run directly:

bash

python main.py

You can change job_url variable in main.py with the job description you want.

⚙️ How It Works

Pipeline Architecture

mermaid

graph TD
    A[Job URL] --> B[Pre-Process Description]
    B --> C{{Cache Check}}
    C -->|Cached| D[Load Candidates]
    C -->|New| E[Fetch from LinkedIn]
    E --> F[Score & Cache]
    F --> G[Generate Messages]
    G --> H[Output Ranked Candidates]

Data Flow

mermaid

sequenceDiagram
    participant User
    participant Cache
    participant LinkedIn
    participant Gemini
    
    User->>Cache: Check existing candidates
    alt Cache Hit
        Cache-->>User: Return cached data
    else Cache Miss
        User->>LinkedIn: Search profiles
        LinkedIn-->>User: New candidates
        User->>Cache: Store with timestamp
    end
    User->>Gemini: Generate messages
    Gemini-->>User: Personalized outreach

🔑Key Components

File Purpose

job_input.py Extracts job details using Selenium

search_linkedin.py Finds candidates via Serper API

score_candidates.py Rates candidates (0-10)

generate_outreach.py Crafts messages with Gemini AI

📊 Scoring Metrics

35% Skills match

25% Experience level

20% Education

10% Location

10% Career trajectory

🚨 Troubleshooting

Cache-Specific Issues:

bash

# Reset corrupted cache
echo "{}" > cache.pkl

# Permission issues
chmod 644 cache.pkl

# Debug cache hits
grep "Loading from cache" search_linkedin.py

Common Issues:

bash

# If Selenium fails:
pip install --upgrade webdriver-manager selenium

# Missing Chrome (Linux):
sudo apt install -y chromium-browser

# API errors:
1. Check quota at https://ai.google.dev
2. Verify keys in .env