Business Context

NovaGraph is hiring an ML engineer for a graph-focused recommendation and risk modeling team. As part of the interview, the hiring panel wants a practical screening model that predicts whether a candidate demonstrates the core technical competencies required for Graph ML work.

Dataset

You are given a structured hiring dataset built from 18 months of interview loops and take-home evaluations. Each row represents one candidate. The target is whether the candidate was rated "Graph-ML ready" by the final hiring committee.

• Size: 12,400 candidates, 30 features
• Target: Binary — Graph-ML ready (1) vs not ready (0)
• Class balance: 28% positive, 72% negative
• Missing data: 12% missing in prior project history, 6% missing in take-home-derived rubric features, and sparse missingness in some interviewer scores

Success Criteria

A strong solution should identify qualified candidates with ROC-AUC e 0.84, F1 e 0.70, and recall e 0.75 on the positive class. The model should also provide interpretable evidence about which competencies matter most.

Constraints

• The recruiting team needs interpretable outputs for hiring calibration
• Batch inference on new applicants should complete in under 200 ms per candidate
• The model should be retrained quarterly as hiring rubrics evolve
• Avoid leakage from final committee notes or post-decision artifacts

Deliverables

1. Build a binary classification model to predict Graph-ML readiness
2. Explain which technical competencies are most predictive
3. Design preprocessing for mixed numeric/categorical data with missing values
4. Evaluate the model with appropriate classification metrics and threshold selection
5. Recommend how the model should be used in production without replacing human judgment