A score that shows its evidence
Four specialist agents read every resume in parallel — skills, experience, culture fit, compensation band. Each returns its own sub-score with the literal span of the resume that produced it, ranked against criteria you set, written back into your ATS.
4
specialist agents per resume, running concurrently
6
stages from a new requisition to a scored shortlist
0
sub-scores without the resume span behind them
Can you explain why a candidate scored what they scored?
A single number with no evidence is not a screening tool. It is a decision nobody can defend.
Every sub-score cites the span of the resume it came from, so a recruiter can audit the reasoning in two clicks and a reviewer can see the basis of a decision months later. Counterfactual bias testing runs on every rubric change rather than once at launch, and the compensation-band agent is deliberately isolated so that what someone was paid before cannot leak into whether they are a fit now.
Five ways screening fails at volume
Recruiters drown in inbound resumes
A high-volume requisition fills the queue faster than anyone can read it, so review order is decided by submission time rather than fit.
Black-box scoring blocks compliance
An off-the-shelf screener returns a single number with no evidence, which is exactly what a hiring decision cannot rest on under transparency rules.
The ATS becomes a dead-end database
It holds the data but cannot reason over it, so the same candidates get re-screened across requisitions and strong past applicants are never revisited.
Generic screeners invent experience
A single-prompt screener will credit a candidate with skills the resume never claimed, and the shortlist collapses on the first phone screen.
Time-to-shortlist loses candidates
The gap between application and first contact is where the strongest candidates accept somewhere else.
From a new requisition to a scored shortlist
The agents run in parallel rather than in sequence, because they are answering independent questions and one slow specialist should not hold up the rest.
The job opens
A requisition is created in your ATS, and a webhook fires the moment the role goes live.
ATS webhooks
Resume ingestion
The candidate record is picked up from the queue within a fraction of a second and routed to the parser layer.
Event-driven
Layout-aware parsing
Structured sections extracted from clean documents, with a fallback path for scans, columns and unusual layouts.
Parsing pipeline
Parallel evaluation
The four specialist agents run concurrently, each emitting a sub-score with the evidence span behind it.
LangGraph · Claude
Weighted aggregation
Rubric weights configurable per role family combine the sub-scores, and low-confidence results are flagged rather than averaged away.
Skills taxonomy match
ATS write-back
Final score, sub-scores, evidence spans and provenance write back to the candidate record, where the recruiter already looks.
Provenance · observability
Six capabilities behind the shortlist
Parallel-agent evaluation
Four specialist agents — skills, experience, culture fit and compensation band — run concurrently per resume, each emitting its own sub-score rather than one opaque number.
Layout-aware parsing
Structured layouts handled directly; scanned PDFs, multi-column CVs and design-heavy portfolios routed to a fallback parser, because a good candidate should not be rejected by their template.
Skills taxonomy matching
Vector search over an embedded skills taxonomy resolves equivalent technologies and adjacent experience, so the match is on capability rather than keyword.
Evidence-backed scoring
Every sub-score ships with the literal resume span that produced it. A recruiter audits the reasoning in two clicks, and legal can see why a decision was made.
Bi-directional ATS write-back
Scored shortlists, evidence spans and audit trails write back to the candidate record in the platform your team already works in.
Bias-audit framework
Counterfactual testing across protected characteristics runs on every rubric change, and the compensation-band agent is isolated so it cannot influence the fit score.
Five pipelines, five different rubrics
High-volume retail and hospitality
Seasonal hiring spikes absorbed without adding recruiters, with the queue worked by fit rather than by arrival order.
Customer support and outsourcing
Large pipelines run through a single system, with multi-tenant write-back across client organisations.
Technical recruiting
Roles where the skills taxonomy matters more than keywords, and adjacent experience should count for something.
Professional services
Licensed-role hiring with credential checks, regional compliance and partnership-track signals built into the rubric.
Healthcare recruitment
Licensure parsing, multi-state credential matching and shift-pattern fit scoring for clinical and allied health roles.
Bring a role you have already hired for
This is our AI engineering practice
It is real work and it is where our four products came from. But what Cognilium leads with is narrower: optimization apps that run in tandem with Microsoft Dynamics 365, computing the decisions the ERP records but does not derive — the optimal price, the optimal pick path, the optimal stock level. See the optimization apps · How we build inside the ERP.