Building Trustworthy AI: Open Source Monitoring in Action

As AI becomes embedded in more and more critical applications, its monitoring remains a blind spot. From data drift and undetected bias to unexpected hallucinations, risks multiply once models are in production. This talk offers a concrete and sovereign approach to AI monitoring, relying on a selection of open source tools such as Evidently, WhyLogs, MLflow, and Truera. Through real-world examples, it will demonstrate how to build a reliable, well-documented monitoring pipeline that complies with the upcoming European AI Act — even for an SME. The goal: to make AI observability both a technical and ethical reflex.