How a staffing firm screens every applicant, minutes after they apply

When a posting goes live, applications arrive faster than any recruiter can read them. The traditional outcome is well known: the first fifty résumés get real attention, the next two hundred get seconds each, and strong candidates who applied on day three are never seen at all. Screening isn't a judgment problem. Recruiters are good judges. It's a throughput problem that quietly becomes a fairness problem.

The shape of it matters more than the volume. Attention doesn't degrade gently across a queue; it falls off a cliff and then flattens near zero. Everyone past that cliff is being evaluated by their arrival time rather than their qualifications.

Nothing about that curve is anyone's fault. It is what happens when a fixed amount of human attention meets a variable amount of inbound. The only way to flatten it is to stop rationing the first pass.

The firm connected its hiring inbox to a Kuhasa agent. From that moment the pipeline runs itself, with each stage handing structured output to the next.

• Intake watches the inbox, recognizes an application, and extracts the résumé from whatever it arrived as: modern PDFs, older document formats, and the occasional résumé pasted into the email body itself.
• Extraction pulls structured history out of unstructured documents: roles, dates, employers, credentials, skills. Formatting variety is the whole difficulty here; two résumés describing identical experience can look nothing alike.
• Evaluation scores the candidate against that specific posting: the job description, the hiring manager's notes, and a set of screening checks the recruiter controls (employment gaps, required experience, education fit).
• Ranking assembles every candidate into one table with a score, a written rationale, and flags on anything a human should look at twice.
• Shortlist drafting writes the presentation email to the hiring manager and holds it for recruiter approval.
• Feedback capture takes hiring manager decisions and folds the reasoning back into how later applicants are evaluated.

A score on its own is not a deliverable; it's an assertion. Every evaluation in the pipeline arrives with the material a recruiter needs to disagree with it:

• the score, and the posting it was scored against
• each screening check, with pass/fail and the evidence that decided it
• extracted work history with the résumé location each item came from
• required-experience assessment stated in the posting's own terms
• employment gaps identified with dates and surfaced without interpretation
• a written rationale in plain language, not a feature vector
• anything ambiguous, flagged as ambiguous rather than resolved silently
• a confidence score, and the résumé itself one click away

Screening tools that promise magic have earned the skepticism they get. Three design decisions make this system accountable rather than magical.

• Every evaluation is evidence-based and inspectable. A score is never a black box; each candidate carries the reasoning, the specific checks that passed or failed, and the résumé itself. A recruiter can disagree, re-assess, or override at any point.
• Nothing irreversible happens without a human. The agent evaluates and drafts; people approve. Shortlists reach hiring managers only after recruiter sign-off, and no candidate is ever auto-rejected.
• The ranking is a starting order, not a verdict. The output is a queue sorted by relevance to this posting. Its job is to make sure the two-hundredth application is read, not to decide who gets hired.

A screening system that scores the same way in month six as it did on day one has learned nothing about the firm it works for. Every posting has unwritten criteria, such as the kind of candidate this client keeps rejecting, the credential that matters more than the job description implies, the adjacent industry experience that turns out to transfer well.

So hiring manager decisions come back in. When a candidate is rejected, that reasoning becomes part of how the next applicants to that posting are evaluated. The recruiter can also adjust screening checks directly and watch the ranking re-order, which is a faster way to encode a preference than describing it in prose.

The honest measure of a screening system is not whether recruiters like the rankings. It is whether the ranking puts the people who actually get hired near the top.

So we score it against outcomes the firm already has: given only what was knowable at application time, where would the agent have ranked the candidates who were eventually placed? That backtest runs on historical postings before the system touches a live one, and it keeps running afterward as new placements land. A ranking that looks reasonable and buries the eventual hire at position 40 is a failure, however good its explanations read.

No candidate is ever auto-rejected. No shortlist goes to a hiring manager without recruiter approval. No email leaves the building unreviewed. The system reads, evaluates, ranks, and drafts, then stops. Every decision that affects a person's application is made by a person.

That constraint isn't only ethical hygiene. Automated employment decisions attract real regulatory attention, and a system that recommends while humans decide is a fundamentally easier thing to operate, audit, and defend.

Today, every applicant to a posting is evaluated within minutes of applying. The two-hundredth application gets the same scrutiny as the first. Recruiters spend their time on the judgment calls the ranking surfaces, not on the reading pile. Time from application to a hiring-manager-ready shortlist dropped from days to 4 hours, across 40 concurrent postings, with 78% of eventual placements ranked in the agent's top 10 on backtest.