Skip to content
KaminClarity for your next step.
Kamin · Trust Center
Trust center

Trust is a product mechanism, not a slogan.

Kamin separates what is operational in the public release from capabilities that remain behind an institutional compliance gate.

Operational

Local processing

Transcript reading and OCR run in-browser; OCR runtime and language files are served from the Kamin origin.

Operational

Local persistence by choice

The default is session-only. With explicit consent, the profile can persist on the same device in IndexedDB, with controls to stop persistence or delete all local data.

Operational

Purpose-specific consent

Transcript analysis and the self-reported profile are separate purposes. Institutional sharing is not operational in the public release.

Operational

Encrypted continuity

Users can download an AES-GCM encrypted local backup and restore it later. Kamin does not store the passphrase and cannot decrypt the backup without it.

AI governance

AI suggests

Extraction or inference does not become fact simply because a model produced it.

Evidence justifies

Signals that affect a judgment must trace to evidence or a governed mapping.

Humans decide

Kamin does not make hiring decisions or block a user from an opportunity.

Safe restore

The portable backup carries evidence, profile state and JSON-LD. On restore, Kamin does not trust stale derived outputs: capabilities and Fit are recomputed from evidence, while advisor/research sharing consents are reset.

What the public release does not do

  • No central student-profile storage.
  • No operational sharing with advisors, employers, or researchers.
  • No use of local profile data to train a central model.
  • Pilot analytics are disabled by default and, when enabled, exclude transcript content and personally identifying academic data.
  • No advertising or commission influences Fit.

Institutional launch gate

Before central storage or cohort analytics can be enabled, Kamin requires at minimum approved hosting, a retention policy, data-sharing agreement, access controls and logging, minimum cell-size rules for aggregation, PDPL review, and re-identification/bias testing.

What will we measure?

Usage analytics are currently disabled. A future university pilot, following required approvals, proposes measuring task completion, time to an understood explanation, correction of input errors and comprehension of recommendation limits. Results will include sample size, dates, methods and errors; growth figures are not published without measurement.

Validation plan · Changelog