UC-CORE-002: AI Nudge Generation (Infrastructure)
Purpose: Generate and deliver AI-driven nudges using LLM with proper guardrails.
| Property | Value |
|---|---|
| Actor | System (AI Engine) |
| Trigger | Scheduled job or real-time event (e.g., low bookings) |
| Priority | P1 |
Main Success Scenario:
- Trigger event detected (e.g., "bookings for tomorrow < 50%").
- Engine retrieves context (historical data, active campaigns) via RAG pipeline.
- LLM generates nudge content (e.g., "Send 20% off promo to recent visitors").
- Guardrails check content for safety, tone, and business logic.
- Nudge is stored and pushed to Owner Dashboard.
Alternate Flows:
- Flow A: Guardrail Failure: If content violates rules, regeneration is attempted up to 3 times before failing silently.
Acceptance Criteria:
- [ ] Nudges generated within defined latency.
- [ ] Bad outputs blocked by guardrails.
- [ ] Nudges correctly linked to suggested actions.