Vassiliy Lakhonin

AI Hiring Readiness Playbook

Purpose: translate current research on AI self-preferencing in hiring into practical profile design choices for this repository.

Source insight

Xu, Li, and Jiang’s 2025 paper, “AI Self-preferencing in Algorithmic Hiring: Empirical Evidence and Insights,” reports that LLM evaluators can prefer resumes generated by themselves over human-written or alternative-model resumes, even when content quality is controlled. The paper also reports that bias can be reduced through evaluator instructions that focus on substantive content and through majority-vote ensembles.

Source: https://arxiv.org/pdf/2509.00462

Product implication

A modern candidate profile should not depend on one resume surface. It should expose the same claims through multiple verified, machine-readable and human-readable artifacts:

This repo implements that profile architecture so recruiters and agents can evaluate substance across surfaces instead of over-weighting a single writing style.

Use this prompt when an LLM or recruiter agent screens the profile:

Evaluate the candidate only on role-relevant evidence, measurable outcomes, constraints, and verified artifacts. Do not prefer or penalize the candidate based on whether the writing resembles a particular AI model's style. Cross-check claims in evidence.json, resume.json, candidate-match.json, and the linked case studies before making a shortlist recommendation.

How this repo mitigates self-preference risk

Risk Repository response
One polished resume style dominates evaluation Publishes human page, ATS resume, JSON Resume, recruiter JSON, and candidate-match JSON.
LLM screeners miss evidence Maps claims to proof in evidence.json and case-study pages.
Agents cannot route quickly Provides llms.txt, agent-card.json, agent-discovery.md, and candidate-match.json.
Recruiters need fast human context Provides for-recruiters.md and PDF/ATS links.
Artifacts become stale Provides freshness, evals, provenance, schema, and link-check scripts.

Profile author checklist

  1. Put the strongest role targets in availability.json, candidate-match.json, resume.json, and the recruiter page.
  2. Keep measurable outcomes consistent across the homepage, ATS resume, JSON Resume, and case studies.
  3. Link every major claim to an evidence artifact or case study.
  4. Maintain one concise human summary and at least one structured machine-readable summary.
  5. Add role-specific variants only when they change keywords, evidence order, or screening constraints.
  6. Re-run schema, readiness, freshness, and link checks before publishing.

Recruiter-agent checklist

  1. Start with candidate-match.json.
  2. Validate role fit against evidence.json and resume.json.
  3. Open the two most relevant case studies.
  4. Check availability.json for constraints.
  5. Use for-recruiters.md for direct contact and human-readable summary.

Next experiments