Job Matching System T2

01 /Overview

An existing job-matching flow used hardcoded rules presented as a decision tree: rigid, redundant, and unable to explain its scores.

02 /What I built

• XGBoost model trained on historical matches in place of hardcoded scoring
• Feature engineering: semantic skill similarity, experience relative to requirements, education hierarchy, certification relevance
• SHAP explanations showing which factors drove each score
• Redesigned flow that computes features once and reuses them
• Documented limitations: training data needs, cold start for new job categories

03 /Outcome

A proof of concept on Streamlit that replaces the rules with an XGBoost model and explains every score, with a written architecture review of the old flow.

04 /What I learned

Explainability is a feature. SHAP values made the model easier to trust than the hand-written rules it replaced.