Raven – Real-time monitoring for ML models

Summary: Raven provides real-time monitoring and alerting for machine learning models by collecting inference logs, detecting data drift and confidence drops, and sending alerts via Slack or email. It stores metrics in ClickHouse and displays them on a dashboard, running entirely within a Kubernetes cluster.

What it does

Raven collects inference data such as confidence, latency, and feature values, detects anomalies like data drift and confidence drops, and sends alerts when issues arise. It installs via Helm and operates fully within the user’s Kubernetes infrastructure.

Who it's for

It is designed for teams running production ML models who need to monitor model behavior in real time.

Why it matters

Raven helps detect and respond to model performance issues promptly without sending data outside the user’s infrastructure.