Case Study 03 · Ranking Engine · Redrob Data & AI Challenge

RecruitRank AI

A hybrid candidate-ranking engine built for the Redrob Data & AI Challenge, producing an auditable top-100 shortlist without relying on keyword matching.

The problem

Keyword-matching resume screens reward buzzword-stuffed resumes over real experience, and they give recruiters no way to see why a candidate ranked where they did. The Redrob Data & AI Challenge asked for a ranking system for a Senior AI Engineer role that was both more accurate than keyword matching and fully auditable — every candidate's score had to be explainable, not a black box.

Approach

  1. Semantic similarity between resume and job description via TF-IDF + TruncatedSVD, with an optional SentenceTransformers backend for higher-fidelity matching.
  2. Career-evidence detection — explicit signals for retrieval systems, ranking, RAG, and production ML experience, not just years of tenure.
  3. Skills and experience assessment layered on top of the semantic match.
  4. Behavioral signals: activity level, response rate, notice period.
  5. Red-flag penalties for buzzword inflation, weak career history, and suspicious patterns in the resume text.
  6. Final score: a weighted blend — 22% semantic similarity, 35% career evidence, 16% skills, plus additional experience/title/behavior/location components, minus red-flag penalties.

Why the core output is deterministic, not LLM-guessed

The whole pipeline is deterministic and CPU-only, with no external API calls. Every component of the final score is a named, weighted term you can trace back to source text in the resume — the opposite of an opaque LLM relevance score.

Tech stack

  • Python
  • TF-IDF + TruncatedSVD
  • pandas
  • SentenceTransformers
  • python-docx

Results

  • Weighted 6-factor scoring model instead of keyword matching.
  • Deterministic, CPU-only pipeline — no external API dependencies, no per-request cost.
  • Produced a ranked, auditable top-100 shortlist for the challenge's Senior AI Engineer role.