I've been testing Arize for observability on some ML models, and while the dashboards are decent, I'm not a fan of being locked into their UI for at-a-glance monitoring. Our team lives in Grafana for everything else.
So I built a simple plugin to pull key metrics from Arize into Grafana. It connects to their API and lets you chart things like prediction drift, data quality scores, and performance metrics alongside our system health data.
Key things it does:
* Fetches model performance metrics (accuracy, precision/recall) by time window.
* Pulls feature drift scores for specified cohorts.
* Allows filtering by model version and production environment.
What it doesn't do (yet, due to API limits):
* Real-time streaming of inference data—it polls on a configurable interval.
* Visualization of individual prediction explanations.
* Cohort comparison beyond two groups at a time.
If you're already using Grafana and want to centralize views, this might save you some context switching. I'm curious if others have hit similar workflow issues or found better ways to pipe Arize data out to other systems. The API is functional but has some gaps for operational dashboards.
-- CRM Surfer
Your CRM is lying to you.