Play 59 — AI Recruiter Agent
Bias-aware AI recruitment — resume parsing with PII redaction (name, photos, graduation years), candidate scoring on skills+experience only (never demographics), explainable factors, job description generation with bias-checked language, paired fairness testing, EEOC 4/5 rule compliance, and always human-in-the-loop decisions.
Architecture
Full architecture details:
architecture.md
How It Differs from Related Plays
| Aspect | Play 60 (Responsible AI) | Play 59 (AI Recruiter) | Play 46 (Healthcare AI) |
|---|---|---|---|
| Domain | General AI fairness | Hiring/recruitment specifically | Healthcare clinical |
| Bias Focus | Any AI system | Employment discrimination (EEOC) | Clinical accuracy |
| PII Type | General PII | Recruitment PII (names, photos, dates) | PHI (HIPAA) |
| Regulation | EU AI Act, NIST | EEOC, Title VII, ADA, ADEA | HIPAA |
| Output | Fairness scorecard | Candidate score + factors + JD | Clinical decision support |
| Key Metric | Disparate impact ratio | 4/5 rule across demographics | PHI recall |
Key Metrics
| Metric | Target | Description |
|---|---|---|
| Matching Accuracy | > 80% | AI score agrees with human recruiter |
| Disparate Impact | > 0.80 | EEOC 4/5 rule compliance |
| PII Redaction | > 99% | Names, dates, photos removed |
| Score Consistency | 100% | Same resume → same score (temp=0, seed=42) |
| Protected Attribute Refs | 0 | Scoring factors never reference demographics |
| Cost per Candidate | < $0.10 | Parse + redact + score |
Cost Estimate
| Service | Dev | Prod | Enterprise |
|---|---|---|---|
| Azure OpenAI | $60 | $450 | $1,800 |
| Azure AI Search | $0 | $250 | $1,000 |
| Cosmos DB | $5 | $120 | $400 |
| Microsoft Graph | $0 | $0 | $0 |
| Azure Blob Storage | $3 | $25 | $80 |
| Key Vault | $1 | $3 | $10 |
| Application Insights | $0 | $25 | $80 |
| Total | $69 | $873 | $3,370 |
Detailed breakdown with SKUs and optimization tips:
cost.json· Azure Pricing Calculator
WAF Alignment
| Pillar | Implementation |
|---|---|
| Responsible AI | PII redaction before scoring, paired bias testing, EEOC compliance, explainable factors |
| Security | Resume PII never reaches LLM, Key Vault for secrets |
| Reliability | Deterministic scoring (temp=0, seed=42), consistent factors |
| Cost Optimization | gpt-4o-mini for parsing, local Presidio redaction |
| Operational Excellence | Fairness monitoring, score history in Cosmos DB |
| Performance Efficiency | Batch candidate scoring, cached JD templates |
FAI Manifest
| Field | Value |
|---|---|
| Play | 59-ai-recruiter-agent |
| Version | 1.0.0 |
| Knowledge | O2-Agent-Coding, T2-Responsible-AI, R2-RAG-Architecture |
| WAF Pillars | security, responsible-ai, reliability, cost-optimization |
| Groundedness | ≥ 85% |
| Safety | 0 violations max |
Last updated on