reliability monitoring

A practical guide to building a production-grade, Linux-native AI uptime pipeline in Bash: a minimal checker that performs real inference, logs success/HTTP codes/latency/GPU stats to CSV, optionally exports Prometheus metrics, and runs every minute via systemd/cron. Includes a reporting script for 24h/7d uptime and latency percentiles, simple alerting, and SLO-focused practices to catch cold starts, GPU pressure, and transient failures pings miss.