A MLOps Platform for All Data Personas
summary
IKEA builds a custom MLOps platform that standardizes seven data personas, moves from MVP to MSP toward a target of 200–300 models, and runs mixed cloud plus on‑prem GPUs to cut costs. Core open source pieces are MLflow, Seldon Core, Feast with Kubernetes, Knative, ZenML, Prometheus and Postgres glued together; deployment supports canary, blue/green and shadow. The platform provides a central feature store and model registry, per‑inference logging for drift detection, and a digital trust score — a functioning, still‑maturing system used in production testing.
AI-written summary · about
chapters (31)
- 0:00Introduction
- 1:02IKEA’s vision and scale
- 4:14Why IKEA needs MLOps
- 7:04Building the MLOps platform
- 9:31Data science personas at IKEA
- 12:18Consumer journeys in MLOps
- 14:43Platform components
- 16:09Development phase
- 17:54Feature store role
- 19:39Deployment and governance
- 23:10Observability and monitoring
- 25:18Model lifecycle
- 26:44User interface
- 29:12Open source choices
- 32:46Conclusion and future
- 33:07Q&A: Kubernetes
- 33:52Q&A: Platform maturity
- 34:56Q&A: Data challenges
- 35:36Q&A: User experience
- 36:22Q&A: Data scientist awareness
- 37:03Q&A: Deployment ownership
- 38:06Q&A: Cost awareness
- 39:29Q&A: Open source plans
- 40:55Q&A: Team size
- 41:36Q&A: Deployment challenges
- 43:01Q&A: Abstraction layers
- 43:45Q&A: Global expansion
- 44:05Q&A: Development pace
- 46:53Q&A: MVP stack
- 49:00Q&A: Real-world data
- 51:07Q&A wrap-up
related talks
53mExperience with an end-to-end MLOps pipeline: Practical insights and 3D architectural perspectives
11mScaling MLOps with Self-Service Platforms
41mTHE STATE OF MLOPS - MACHINE LEARNING IN PRODUCTION AT ENTERPRISE SCALE
13mMLOps at Scale
50mMLOps for a hybrid AI platform compliant with European law
33mBuilding Resilient (ML) Pipelines for MLOps
42mMLOps 2025: A Journey into the Past and the Future
41mFrom Prototype to Production Building a Scalable MLOps Platform from a Reference Arch