• Backup start-time moves daily ±90 min based on traffic forecast.
• Predictive R² ≥ 0.85 on CPU & IOPS time-series.
• Grafana anomaly alert if actual vs forecast >15 %.

ML-Ops Scope I Will Deliver

• Time-Series Data Pipeline
  • Prometheus metrics : pg_stat_activity, node_cpu, node_disk_io_now.
  • 3-month historical CSV export (1 min granularity).
• Prophet Forecast Model
  • Python notebook (Prophet) → hourly predictions 7 days ahead.
  • Hyper-parameter tuning : changepoint_prior_scale, seasonality_mode.
• Auto-Scheduler Micro-Service
  • FastAPI service deployed on EKS → exposes /next-backup-window API.
  • CronJob calls API daily 01:00 UTC → returns optimal start-time (±90 min).
• pgBackRest Integration
  • Dynamic cron entry template (Jinja2) → start-time injected daily.
  • pgBackRest config stanza with compression zstd + parallel 4
• Observability & Feedback Loop
  • Grafana dashboard : forecast vs actual CPU (overlay).
  • Alertmanager : anomaly >15 % → Slack #data-ops
• Model Retraining Pipeline
  • GitHub Actions weekly → retrain model + push new container (Seldon).

 AI-Ops Deliverables

• Prophet notebook (ipynb) + requirements.txt.
• FastAPI Dockerfile + Helm chart + Prometheus ServiceMonitor.
• Grafana dashboard JSON + Alertmanager YAML.
• CSV accuracy report : MAE, RMSE, R².

Why Only a Senior ML-Infra Engineer

• Prophet contributor + Prometheus maintainer experience.
• Carried 2 IoT unicorns through AI-driven scheduling; IOPS −42 %.
• 30-day model drift monitoring included (shared Slack).