/* case study */
Self-Healing Content System
Detects when upstream Google Cloud docs change, traces which GCP certification practice questions are affected, and proposes AI fixes for human review.
// impact: Change detection → provenance impact analysis → AI healing → human review
// Context
Practice questions for the GCP Professional Cloud Architect exam are written from Google Cloud documentation. When that documentation changes, questions quietly go stale.
// Challenges
- ›Find out cheaply when any of many upstream sources has actually changed
- ›Work out which questions depend on which source passages
- ›Let AI propose fixes while a human stays in control of what ships
// Approach
- ›Cache-aware source polling with Vercel Cron and snapshot versioning
- ›Provenance links from each question to its sources, used for impact analysis and triage
- ›A healer agent (Vercel AI SDK + OpenRouter) that drafts proposed edits
- ›An approve / reject / edit review queue, plus a 'Cloud Next' simulation with live SSE progress
// Outcomes
- ›A complete detect → triage → heal → review loop, deployed
- ›Dashboard, sources, changes, review-queue and content views