This repository contains the JOYBOY_07 Ranker, a multi-stage cascading pipeline engineered for the Redrob Talent Intelligence Challenge.
Our engine is designed to operate under strict CPU-only constraints while aggressively filtering for dataset traps.
- Deterministic Guardrails: Hard-filters honeypots and "keyword stuffers" (e.g., non-engineering titles with high skill keywords).
- Hybrid Retrieval: Uses optimized TF-IDF sparse matching to prune 100k candidates to the top 2k, followed by dense semantic scoring for relevance.
- Behavioral Calibration: Applies a mathematical multiplier to prioritize candidates with high recruiter response rates and recent platform activity.
- Backend: Python, Scikit-Learn (TF-IDF), Pandas, NumPy.
- Frontend: Streamlit with custom Glassmorphism CSS and Plotly-integrated visualizations.
- Performance: Optimized for <5 minute execution on CPU-constrained environments.
- Clone the repository:
git clone https://github.com/07MADARA/talent-ranker - Install dependencies:
pip install -r requirements.txt - Run the ranker:
python rank.py(ensurecandidates.jsonlis in the root directory). - Validate results:
python validate_submission.py team_antigravity.csv
Submission by Team JOYBOY_07