How LY Corporation Uses MLflow: An Overview

mlflow

LY Corporation, a major Japanese tech company, shares how they built and operate a managed MLflow service for their internal AI platform supporting 100+ products. Key topics include centralizing MLflow on Kubernetes with per-service isolation, implementing service-to-service authentication using OAuth 2.0 Client Credentials Grant Flow with Athenz and mTLS-based identity via SPIFFE certificates, and enforcing authorization through an Authorization Proxy sidecar without modifying MLflow's codebase. The post also covers monitoring with Prometheus and usage trends showing a fourfold increase in API requests over 18 months, with MLflow increasingly used for logging evaluation metrics beyond just storing training artifacts.

Enterprise-Scale MLflow Operations and Security Practices at LY Corporation

How LY Corporation Uses MLflow: An Overview ​

Implementing Service-to-Service Authentication and Authorization Based on OAuth 2.0 ​