Product overview

One verification platform. The whole hiring lifecycle.

Census orchestrates your existing hiring stack (background checks, deepfake detection, behavioral biometrics, OSINT) into a single, explainable risk score that travels with a candidate from application through continuous post-hire monitoring.

What Census is

The orchestration layer that point tools lack.

Census is the orchestration layer that point tools lack. We sit behind your existing background-check, identity, deepfake-detection, and behavioral analytics providers and reason across them, so the joint pattern that no individual tool can see becomes visible. The output is a single trust score per candidate, with the underlying evidence attached, updated at every meaningful moment of the hiring lifecycle.

One score per candidate

Built to be acted on directly.

Census produces a single risk score, mapped to plain operational decisions: proceed with hiring, route for review, or pause for additional verification. Every score decomposes into the evidence behind it, and every borderline case is surfaced to a human reviewer with the full evidence bundle attached.

Four signals, jointly reasoned

Fooling one detector is not the same as fooling the system.

Census fuses four independent signal classes. The platform does not depend on any single one.

01
Pillar one

Resume Intelligence

Claims are extracted from application materials and cross-referenced against external sources: education, prior employment, certifications, projects, and identity provenance. The output is a per-category trust breakdown. For example, three of four employment claims verified at high confidence, while one education claim could not be matched to the source institution.

That granularity matters when a candidate is rejected: regulators and the candidate themselves are entitled to an explanation specific enough to be challenged.

02
Pillar two

Video Authenticity

A multi-provider deepfake and liveness ensemble runs across video, image, and audio streams during interviews. No single detector is robust against every face-swap toolchain; the ensemble is the defense.

Census surfaces an authentic / suspicious / deepfake-detected verdict with the contributing evidence (including frame-level disagreements) attached.

03
Pillar three

Behavioral Biometrics

Per-user computer behavior baselines are established during onboarding and tracked thereafter, with deliberate tolerance for legitimate drift (new hardware, fatigue, skill development). The pillar catches the cases point tools cannot: account sharing, operator substitution after a clean hire, or quiet credential transfer.

Paired with insider-threat tools, behavioral biometric scores are generated continuously.

04
Pillar four

Historical Consistency

Cross-session synthesis over a rolling multi-year window. Are the candidate's stated facts stable across applications? Is the video authenticity score holding up across interviews? Are the behavioral patterns consistent, or do we appear to be looking at different operators behind the same identity?

This is the pillar that catches the post-hire pivot.

The full hiring lifecycle, scored

Most verification tools fire at one moment and disappear. Census scores at four.

01

Application

Resume and document checks at the moment a candidate enters your funnel.

02

Finalist

Finalist interview, video and document signals.

03

Offer

Full multi-pillar score, immediately before credentials are issued. The most consequential single check.

04

Continuous post-hire

Behavioral and historical signals on a rolling cadence; document and video pillars rerun on triggered events such as a transfer to a higher-clearance role.

The moment of hire is no longer a hand-off out of risk visibility. That is the failure mode that has let foreign IT-worker schemes draw US salaries undetected for months. Census closes it.

Explainable by design

Every Census decision carries the evidence that produced it.

Per-claim breakdowns, per-pillar voting, the threshold actually used, and any adaptive adjustments are all preserved on the verification record and retrievable through the audit log. When a candidate is rejected, your team can produce an explanation specific enough to be challenged on its merits, not "low score, rejected."

Human-in-the-loop on every borderline call

Census is advisory. No candidate is automatically rejected on a single low score.

Scores route to a designated reviewer with the full evidence bundle attached, and the reviewer's decision is recorded alongside the platform output. Census is built so a wrongful rejection is recoverable through human review, and the platform errs toward caution when signals disagree.

Adversarially robust

Caution under uncertainty is the default.

Inside each pillar, Census uses heterogeneous detectors: multiple deepfake models with different architectural lineages, multiple OSINT sources with different data-acquisition methods. An attacker engineering against one architecture is not automatically engineering against the others. When pillars disagree, the platform is designed to route for review rather than auto-clear.

Plays nice with your stack

Invoked by your ATS and HRIS at each lifecycle event.

Greenhouse, Lever, Ashby, iCIMS, Workday, BambooHR, ADP and others integrate via the platform's API and webhook callbacks. Behind the scenes, each pillar is built behind a provider abstraction, so your background-check, deepfake-detection, or behavioral biometrics vendor can be swapped out without changing your integration. Verification events stream to your SIEM via structured logs.

Built for the data your security team owns

Tenant isolation enforced at the database level.

PII is encrypted at rest with tenant-scoped keys. Organization data does not commingle. Retention windows are configurable per data category, with automatic expiration. Behavioral baselines and raw media are handled with explicit candidate consent and short default retention. Raw biometric data does not leave your tenant.

What Census is not

Get started.

A short conversation, then a real candidate workflow.

Become a development partner Try it now