[1.9] Document refresh process (knowledge base currency)

[jasminetay-moe]

Epic: Part 1 — Technical Integration
Sprint: Sprint 3
Priority: P0
Role: Engineer

User Story

As an Engineer, I want to set up a document refresh process so that the knowledge base stays current when domain owners update source materials.

Context

Domain owners will upload updated guidance materials via the Management Portal (Part 6, story 6.5). The refresh process is the backend mechanism that detects those changes and re-ingests updated documents into the vector store. These two stories form two halves of the same loop — the trigger interface between them (e.g. webhook, queue, scheduled scan) must be agreed before either is built.

Acceptance Criteria

• Refresh trigger mechanism defined and agreed with the 6.5 engineer (e.g. webhook on upload, scheduled scan, manual trigger)
• Refresh process detects new, updated, and removed documents in cloud storage
• Re-ingestion runs correctly on changed documents (or full re-index if needed)
• Stale embeddings removed when source documents are deleted
• Process can be triggered manually by an engineer or PM as a fallback

Dependencies

• Blocked by: [1.8] Chunking + embedding pipeline (Vertex AI Vector Search) #13 (1.8) — chunking + embedding pipeline
• Coordinate with: [6.5] Build knowledge base management UI (upload, tag, delete) #50 (6.5) — knowledge base management UI (the upload trigger that initiates refresh)

---

📄 PRD: Part 1 — Glow CI PRD