Project Record

HR Recommendation Algorithm

Candidate ranking engine for quant hiring using weighted skill vectors and behavioral fit signals.

HRrankingquant-recruiting

Context

Hiring for quant teams requires balancing technical depth with collaboration and execution behavior. Resumes and profiles are high-dimensional and noisy; a ranking engine that combines hard skills, culture fit, experience, and practical constraints helps shortlist candidates consistently.

Problem

We need a single score that reflects (1) match to role requirements (skills, tools, domain), (2) behavioral/culture signals from assessments or past feedback, (3) experience level and relevance, and (4) optional factors like location or availability. Inputs are sparse and come from different sources (CV, assessments, internal tags).

Model

We score candidates with a weighted blend:

S = 0.55 \cdot f_{skills} + 0.25 \cdot f_{culture} + 0.15 \cdot f_{experience} + 0.05 \cdot f_{location}

$f_{skills}$ is typically cosine similarity between candidate and role skill vectors; $f_{culture}$ from survey or interview scores; $f_{experience}$ from tenure and relevance; $f_{location}$ binary or distance-based.

Implementation

Skill vectors: bag-of-skills from job description and candidate profile; optional TF–IDF or embeddings for richer matching.
Culture: normalized scores from internal rubrics or assessment tools.
Ranking: sort by $S$ ; optional diversity or constraint filters (e.g. max N per school).

Trade-offs

Cosine similarity is robust for sparse skill vectors, but the quality of culture signals depends on input consistency. Weights can be tuned per role or team. The model is interpretable so recruiters can override or adjust when the score conflicts with domain judgment.