Practical MLOps- Do we need all the things
summary
Claim: a small subset of MLOps delivers real production value rather than every platform piece. Example: a reproducible pipeline built with Kedro catalogs and pipelines, an isolation forest anomaly detector on a temperature dataset, and MLflow model tracking. The repo packages into a pip-installable project, logs datasets and intermediate artifacts, and reruns end-to-end, demonstrating reproducibility, modular nodes, lightweight packaging, and that you can maintain deployed models without full-featured feature stores or heavy infrastructure.
AI-written summary · about
chapters (13)
- 0:00Practical MLOps intro
- 1:50Why traditional ML fails
- 3:59MLOps rise and challenges
- 13:11MLOps pipeline components
- 15:18Kedro project setup
- 21:16Kedro catalog for data
- 28:06Kedro pipelines and registries
- 33:20Running and logging a complete ML pipeline
- 41:02Reproducibility and modularity
- 47:43Framework for long-term projects
- 53:08Testing embedded ML systems
- 1:09:08Do we need all MLOps tools?
- 1:12:16Q&A
related talks
25mBuilding Scalable MLOps Pipelines That Actually Work
36mIs the MLOps tooling the same as the DevOps tooling?
13mMLOps: From AI Research to Production-Ready Systems
14mWhy Your DevOps Skills Are Your MLOps
40mA Hitchhiker Guide to MLOps
33mBuilding Resilient (ML) Pipelines for MLOps
47mMachine Learning in Production: An Intro to MLOps.
53mExperience with an end-to-end MLOps pipeline: Practical insights and 3D architectural perspectives