MLOps at Renault: A Generic Pipeline for Scalable Deployment
summary
Claims a Python-first Kubeflow + GCP Vertex AI pipeline that deployed 12 supply‑chain models in months, now used by 16 projects with 5 in production and roughly 100 DSs in scope. It provides a repo template, one‑model‑per‑repo, CI/CD that moves from manual clicks to automated unit and end‑to‑end tests, prebuilt Docker images, drift monitoring and automatic retrain triggers. Pitfalls include shared repo complexity and dependency hell, fixed by code duplication to remove external deps and one repo per model and clearer CI.
AI-written summary · about
chapters (13)
- 0:00MLOps journey at Renault
- 1:51First large-scale ML deployment
- 4:00Vertex AI pipelines for supply chain
- 7:09Scaling MLOps across Renault
- 10:24MLOps as a service
- 12:32Conway's law pitfalls
- 16:26Rebuilding architecture
- 18:59Current pipeline and demo
- 20:26Data scientist workflow
- 22:46Monitoring and alerting
- 23:49Key lessons
- 25:16Future plans
- 25:58Q&A
related talks
25mBuilding Scalable MLOps Pipelines That Actually Work
23mAccelerating Model Deployment with MLOps
26mKubernetes and MLOps for Scalable and Reproducible Generative AI
24mFrom Model to Production: Automating the ML Lifecycle with CI/CD
54mScaling Machine Learning Pipelines in Cloud
49mKubernetes and MLOps for Scalable and Reproducible Generative AI
26mBuilding Portable ML Pipelines for Development and Production
10mFrom Model to Production: Building Scalable MLOps Pipelines