FeatInsight: An Online ML Feature Management System on 4Paradigm Sage-Studio Platform

Feature management is critical for online machine learning (ML) applications but often becomes the performance bottleneck, contributing up to 70% of the total latency in services like sales prediction. Poor feature configurations (e.g., excessive irrelevant features) can degrade model generalization. However, managing online ML features is non-trivial due to (1) large-scale, heterogeneous raw data (e.g., 17 tables with hundreds of columns in the 2018 PHM dataset), (2) the need for consistent, low-latency computation over interdependent features, and (3) real-time update and deployment requirements. In this demo, we present øursys, a unified system for managing the complete feature lifecycle, covering feature design, storage, visualization, computation, verification, and lineage. Powered by OpenMLDB, øursys has been deployed in over 100 real-world scenarios on 4Paradigm's Sage Studio platform, handling trillion-dimensional feature spaces and enabling millisecond-level updates. We demonstrate how øursys boosts feature design efficiency (e.g., in product recommendation) and accelerates computation performance (e.g., in fraud detection). Code is available. The system is available at: https://github.com/4paradigm/FeatInsight and a demonstration video can be found on YouTube at https://youtu.be/GS9ZCJhtzyQ?si=7O8-dGQS84x07cTM.