milestones
M0 GPU Foundation in progress Tesla T4, DCGM, Docker GPU passthrough
M1 GPU Metrics Exporter queued NVML → Prometheus, custom Go binary
M2 DCGM + Observability Stack queued Grafana dashboards, Alertmanager rules
M3 Inference Layer + Token Path queued vLLM + LiteLLM + Open WebUI, TTFT dashboard
M4 - M8  ·  SLOs, load testing, chaos, postmortems, OSS cadence