TIER A ⚖ ANNEX III HIGH-RISK BLUEPRINT

Hiring Screening with Bias Controls

Screen candidates faster without an AI Act or discrimination liability - blind by construction, bias-audited, and a human behind every decision.

from €1,990

CV INTAKE - BLIND SCREENING · LIVE

Candidate #217 - raw CV

NAMEVardenis PavardenisMASKED

AGE52MASKED

PHOTOportrait.jpgMASKED

EXPPython · MLOps · logistics domain

LANGLT · EN

⚑ protected attributes masked before any scoring

cand #218 ▸ match 0.41 · borderline → HUMAN REVIEW · no auto-reject

Evidence - job-related criteria only

Python · 6 yrs ⌕ CV §2 MLOps · 3 yrs ⌕ CV §2 domain: logistics ⌕ CV §3 LT + EN ⌕ CV §4

match 0.87 · every point traced to evidence

blind screening ✓ criteria job-related ✓ adverse impact 4/5th 0.93 ✓

🟡 SHORTLIST SUGGESTED - recruiter decides. The AI never rejects anyone.

batch ▸ 100 CVs · 34 suggested · 0 auto-rejected · all decisions logged ⛓

How it works

Define job-related criteria

Designed with your recruiters, each criterion documented as job-related - the legal test a discrimination claim will apply.

Blind the pipeline

Name, age, gender, photo, nationality masked before scoring. The model cannot discriminate on what it never sees.

Score with evidence

Skills extracted with citations to the CV; match score with reasons. Borderline candidates route to human review - never auto-reject.

Monitor & document

Adverse-impact dashboard on outcomes, full decision log, candidate transparency note. Audit-ready by default.

The Four Guarantees™ - this build

Measured value

~70% screening-time cut at reference volume; eval gate: extraction accuracy + rank agreement with your senior recruiters on a blind set.

Defensible

Blind by construction; every score traced to CV evidence; Annex III obligation map + decision log; 4/5ths adverse-impact tests documented.

Self-correcting

Adverse impact monitored monthly on outcomes, not just scores; recruiter overrides feed calibration; drift alerts on candidate mix.

Yours & everywhere

Runs in your cloud, wired to your ATS, full source. MCP endpoint for recruiting copilots - under the same controls.

The number, sized honestly

Reference buyer: Lithuanian or Baltic employer / staffing agency screening ~200-1,000 CVs per month, with recruiters spending 5-10 minutes per CV on first pass.

~70% first-pass screening time cut - the agreed number (conservative)

0 auto-rejections - every "no" is a human's decision

4/5ths adverse-impact rule monitored monthly on real outcomes

~2 mo payback at the PoC tier

Three ways to own it

Tier	What you get	Price
Scaffolding ★	The full repo - blind-screening pipeline, evidence extractor, adverse-impact harness, Annex III obligation checklist, MCP server. Reference run on synthetic CVs.	€1,990
PoC ★★RECOMMENDED	Run on your historical CVs and decisions (bias-checked first) - extraction accuracy, rank agreement and adverse impact measured. Code + quality report yours. High-risk domain: evidence before any promise.	from €6,000
Implementation ★★★	Production: ATS integration, blind pipeline + human gates enforced, adverse-impact monitoring, retraining under review, AI Act documentation maintained - the agreed number (conservative) attaches here.	from €18,000

★ = engagement depth. PoC is the recommended path: quality proven on your data before production money. The PoC carries no performance guarantee by design; the agreed number (conservative) attaches at Implementation, informed by the PoC report.

What we don't promise

We won't sell you auto-rejection. Under the AI Act this is a high-risk system, so the software suggests and documents - humans decide, on every candidate. We also won't promise "bias-free": we promise measured, monitored, and documented, with the adverse-impact numbers reported to you monthly. A vendor promising fully automated, bias-free hiring is selling you a liability.

Ready to see your own number?

Request the build: within 48h you get a personal reply with the value sized to your volume.

No commitment · reply within 48h · your data stays in the EU