Changing KYC rules shouldn't mean rebuilding onboarding.

How it works

1. Step 1

   Write the policy

   An admin describes the rules in plain English. Each sentence is mapped to one of nine supported rule kinds; anything that can't be mapped is flagged, not guessed.

2. Step 2

   Customer applies

   A guided journey built from that policy: consent, document upload with instant quality checks, and a selfie.

3. Step 3

   AI proposes, with evidence

   Each finding records the document and field it relies on, and code computes a transparent risk score. In mock mode the readings come from synthetic ground truth.

4. Step 4

   A human decides

   Clear cases can be auto-approved within the policy's threshold. Nothing is rejected automatically: integrity issues go to a person.

Built to be trusted

What this demo does today, and what is still to come.

• Synthetic data onlyLive

  Fictional specimen documents with ground truth, and a demo notice on every page.

• Human final decisionLive

  The pipeline never rejects on its own; integrity flags always route to manual review. A reviewer console is coming.

• Audit eventsLive

  Every pipeline step and consent emits an audit event. In this demo they go to the server log; database storage is coming.

• Access controlsLive

  A demo access code, a per-IP rate limit and a daily cap on AI calls guard every route that uses AI or changes policies.

• Encrypted per-customer storageComing

  A Vault Transit helper with one key per customer is built and tested. The app does not store uploads yet.