Skip to main content

AI Integration

:::info Source Sourced from services/certification-service/AI_INTEGRATION.md in the documentation repo. :::

1. Scope​

Minimal direct AI. Two touchpoints:

  1. AI-generated certificate template suggestion (authoring assist, M4+).
  2. Automated revocation detection (M5+): flag suspicious issuance patterns.

2. Template Suggestion​

  • Prompt certification.template.suggest — input: tenant brand, course subject, audience; output: layout JSON + color palette.
  • Via AIClient port; provenance on generated template; HITL review required before activation.

3. Revocation Detection (M5+)​

  • Anomaly detection on issuance patterns (e.g., mass issue from one admin in short time → flag for review).
  • Input: recent issuance metrics; output: confidence score + flag.
  • NOT auto-revoke; flagged to compliance officer.

4. Safety Requirements​

  • No PII to AI for template design.
  • Revocation detection uses only aggregated metrics (counts, times, course IDs) — no learner PII.

5. Provenance​

  • AI-suggested templates track prompt version + trace.
  • AI flags on revocation detection persisted with evidence + confidence.

6. Classification​

  • Template suggestion: limited-risk.
  • Revocation anomaly detection: advisory, not decision-making → limited-risk.

7. Caching​

  • Template suggestions cached by (brand+subject+audience) hash, 30-day TTL.

8. Local vs Cloud​

  • Template suggestion: cloud (quality).
  • Anomaly detection: cloud + batched.

9. Right to Explanation​

  • AI-flagged revocation candidates include rationale displayed to compliance officer.