HyperVerge

Every function is a model, from face matching and passive liveness to document reading and forgery detection, which fits a company that began as a computer vision team. Remove the models and no marketplace, orchestration engine or rules layer is left.

Agent assisted video KYC keeps one human step, with real time dashboards, beside an agentless option, and everything else runs without a person by design. An applicant the automated path cannot verify meets no described review queue, escalation threshold or alternative evidence route.

Independent laboratory testing stands behind liveness and deepfake detection, which conforms to ISO 30107 3 Level 2 for presentation attack detection. Docs AI is said to catch forged documents with more than 95 percent accuracy, but the figure has no method or test set, and the lending layer that feeds credit decisions is unmeasured.

Named customers with measured outcomes are published in a case study library. The homepage adds self reported aggregates of more than 500 enterprises and over 2 billion verifications.

Model quality in computer vision rises with training breadth, so the pull to pool applicants' images across institutions is strong. Whether one institution's submissions train models serving another is never addressed in published material.

Service data falls under separate terms that are not published, since the privacy policy covers the website alone. That data includes biometric captures, identity documents and, in lending, screening against court and police record databases, with no stated limit on how long a facial template persists.

ISO 27018 for personal data in the cloud is the recognized certification on the badge row, shown without certificate or scope. A SOC 2 badge sits beside it with no report type, and GDPR and FATF appear there too, though both are regulations or standards rather than certifications.

Products are built to named rules: the Reserve Bank of India's video customer identification process and, for the Philippines, the Anti Financial Account Scamming Act with three central bank circulars. Building to a rule is not admission, though, and a government procurement award shows capability rather than status.

Face matching performs unevenly across skin tone, age and gender, which tests the homepage promise to verify every demographic without bias. HyperVerge publishes no per group error rate or method and names no independent demographic test such as the NIST face recognition evaluations.

Auditability and control are said to be built in, yet no audit trail, rejection reason or contest route is described. Its inclusion mission pulls against wrongful rejection, but HyperVerge offers no published guarantee, and an applicant whose face fails to match is not told which system decided.

The computer vision stack is built in house and covered by HyperVerge's own patents, US Patent 12,633,162 among them. Data comes from identity databases, business registries and credit bureaus named only by category, with no specific provider identified.

Inputs are well covered: APIs, mobile and web SDKs and a no code path, with documents from India's national identity number to passports and driver's licenses. Downstream, no core banking or loan origination system is named, so how a result reaches staff is unclear.

Website data sits on servers in India and with HubSpot in the US, per the privacy policy, but the service itself has no published hosting provider, region or residency commitment. With biometric captures moving across several transfer regimes, that gap matters.

A consultative sale built around each client's problem stands in for a price list, and every route in is a demo request. Charges could fall per verification, per active customer or by subscription, and nothing published says which.

Banks, brokerages and insurers each get their own proposition, with a lending layer that reaches underwriting documents, and reach runs from India to the Philippines and Nigeria. Financial services is still one vertical among several, beside telecommunications and gaming.