Jev Resume Screening for Small Businesses: The Service, and the Legal Risk First

You are not going to read 240 resumes properly. Neither is your client, and neither is whoever they currently pay to try. The average job posting now draws 240+ applications — about 3x the 2017 level — which is exactly why small businesses and independent recruiters reach for AI resume screening in the first place. Jev resume screening can do that job well: score and rank a batch of resumes against objective, job-related criteria in minutes instead of an evening.

But there’s a catch, and it’s the reason this page exists. Screen resumes with AI badly — auto-rejecting people with no human review — and you’re not saving your client time, you’re handing them an EEOC disparate-impact problem with their name on it. Can you make money running Jev-powered resume screening? Yes — but on this page, the legal risk has to come before the pitch, not after it, because getting this part right is also the actual selling point. This page covers the compliance issues first, then how to build the service.

Input → output: what a responsible screening pass looks like

Say a client hands you 40 resumes for a “Senior Backend Engineer” role with one hard requirement: 5+ years of backend experience, plus a nice-to-have (AWS certification). Illustrative example, not real applicant data:

Input: 40 resumes (PDF/text), criteria = years_backend_experience >= 5 (hard filter), has_aws_cert (scored, not filtering).

Output (Jev returns a ranked list with reasoning per entry):

Rank	Candidate	Years (backend)	AWS cert	Score	Reasoning
1	Candidate #17	9	Yes	94/100	Meets hard filter; cert present; explicit backend title history
2	Candidate #03	7	Yes	88/100	Meets hard filter; cert present; some frontend overlap
3	Candidate #29	6	No	71/100	Meets hard filter; no cert; strong tenure at relevant employer
—	Candidate #11	3	Yes	excluded	Below hard filter (5 yrs); flagged for optional manual re-review
…	(36 more rows)

That’s the entire deliverable: a ranked list plus the “why” behind each score — not a rejection list, and not a hire decision. The client’s own people still decide who gets a call. That hand-off is what the rest of this page is about.

The risk, first: AI resume screening isn’t just a technical problem

As of the current regulatory environment (see references below), a few things are clear:

The employer bears full liability, and can’t transfer it. Even if you built the screening tool, the company using it stays liable under Title VII for the results — that protects you, but it also means clients will care a lot about getting this right, and will hold you responsible if something goes wrong with what you built them.
No fully-automated adverse decision. Guidance clearly calls for a qualified human to review an AI recommendation before a rejection, with the authority to override it, and a documented review trail.
Criteria must be directly job-related. You can’t use proxy variables correlated with protected characteristics (race, gender, age, etc.) as a basis for judgment, even ones that look neutral on their face.

This page does not recommend building an “auto-reject” feature — even if it’s technically doable, it’s the single riskiest selling point available here. The right positioning: Jev ranks/scores resumes against objective criteria; the final decision and all communication stay entirely with a human.

Evidence: the community is building this, but not always the right way

typesafe-jev — screen a folder of CVs with the Jev decision model: typed judgments, an editable policy, free re-scoring.
jev-resume-analyzer — CV diagnostics and job-alignment scoring, built with React and FastAPI.
jev-resume-disqualifier — self-described as an “automated resume knockout engine” that claims to “eliminate 80% of unqualified applicants” and describes itself as “EEOC-safe.” This project’s name and positioning are themselves a cautionary example: an open-source tool calling itself compliant doesn’t make it compliant, and “automated elimination” is exactly the high-risk pattern described above. We’re listing it as evidence that this demand exists, not as an endorsement of its product design.

How to build a responsible version of this service

What it includes: score and rank a batch of resumes against objective criteria (a hard years-of-experience requirement, specific skill keywords, certifications) for a small business or independent recruiter, delivering a ranked candidate list with the reasoning behind each score. Whether to interview or reject stays with the client’s own people.

Who buys it: small businesses without dedicated HR, or independent recruiters, who don’t want to hand-review every resume but also don’t want to hire full-time help.

Pricing reference: charge per batch — for example $99–$299 per 50–200 resumes (extrapolated from batch-work pricing in our Muse errand-service playbook, not a verified market rate). We don’t recommend a monthly “auto-reject” subscription model — that’s exactly where the risk concentrates.

Deliverable: a ranked candidate list, the scoring rationale for each entry, and a disclaimer stating clearly that this is a ranking recommendation, not a hiring decision, and the final call rests with the client.

A compliance recommendation to give your client (this is itself part of your service’s value): tell them the AI ranking needs human review, can’t be the sole basis for rejecting a candidate, and that review needs to be documented. You’re not covering yourself by saying this — you’re helping the client avoid real legal exposure, and that’s exactly the kind of advice that builds trust.

FAQ

Who’s actually liable when AI screens resumes? The employer — not you as the service provider. The EEOC has made clear that even when a screening tool is built or run by an outside vendor, the employer stays fully liable for the results under Title VII, and can’t contract that liability away to a software developer. That doesn’t mean you’re off the hook, though — if your service is designed irresponsibly (say, encouraging fully automated rejection), and a client gets into trouble, you’re the first person they’ll remember.

Why does this page argue against building an “auto-reject” feature? Because a fully automated adverse decision with no human review is exactly the risk pattern regulators focus on — EEOC guidance calls for a qualified human to review AI recommendations before any adverse decision, with the authority to override it, and documented evidence that review happened. “Rank candidates for a human to review” is not just a safer way to phrase this — it’s a real compliance boundary, not a marketing choice.

Is Jev actually well-suited to this task — what is it good at here? Jev is good at typed judgments: does this resume meet an objective, bounded criterion (years of experience, a specific certification), or how does it rank on some dimension. It’s not suited to — and shouldn’t be asked to make — subjective judgment calls like “culture fit,” which are exactly where the legal risk concentrates. The more objective and job-related the criteria, the lower the compliance risk.

Are there real paid case studies for this? We didn’t find any. The projects referenced below are community open-source tools, not verified paying-client case studies — worth stating plainly here more than in any of our other Jev pieces, because honest disclosure matters more on this particular topic.

References

gcheck.com — “EEOC Compliance Requirements 2026: Complete Employer Guide to Fair Hiring Practices” — employer liability, disparate-impact framework.
Warden AI — “EEOC AI Hiring Guidance: Key Employer Compliance Rules” — human-review requirements, documentation requirements.
Metaintro — “Ghost Jobs Are Breaking the 2026 Hiring Market” — applications-per-posting volume figure (240+), also cited in our job-search-intel skill piece.
typesafe-jev
jev-resume-analyzer
jev-resume-disqualifier — listed as demand evidence, not an endorsement of its product positioning; see caveat above.
TypeSafe AI