Raven is a tool for real-time monitoring and alerting for ML models — something like Grafana, but for your model’s behavior instead of infrastructure What it does: * Collects inference logs (confidence, latency, feature values) * Detects data drift and confidence drops * Sends alerts to Slack / email when something goes wrong * Stores metrics in ClickHouse and shows them in a clean dashboard It installs with a Helm command and runs entirely in your own k8s cluster (no data leaves your infra).