The State of AI and Machine Learning Operations in 2019
summary
Overview of 2019 MLOps focusing on reproducibility, orchestration and explainability with concrete tools and demos: reproducible pipelines using Kubeflow-style pipelines, DVC and Pachyderm with persistent volumes storing inputs/outputs; model orchestration via Seldon containers, GitOps integration and PyTorch Hub dynamic loading (deployed as REST/gRPC services with Prometheus metrics); an NLP pipeline (Reddit hate speech + Brexit stream) and an income-classifier + Alibi anchor explainer deployed to Kubernetes that return tokens, predictions and human-readable explanations, demonstrating end-to-end runs and logs.
AI-written summary · about
chapters (12)
- 0:00Start
- 0:44SMALL DATA SCIENCE PROJECTS
- 4:17ABSTRACTING INDIVIDUAL STEPS
- 5:04CASE STUDY
- 9:58PACHYDERM
- 12:29ADDING GOVERNANCE/COMPLIANCE
- 13:27STANDARDISATION OF METRICS
- 14:46STANDARDISATION OF ERRORS
- 15:09COMPLEX DEPLOYMENT STRATEGIES
- 15:42GITOPS STRATEGIES FOR ML
- 22:56EXAMPLE
- 26:59DESIGN PATTERNS WITH CRDS
related talks
35mThe State of AI and Machine Learning Operations in 2019
26mKubernetes and MLOps for Scalable and Reproducible Generative AI
53mExperience with an end-to-end MLOps pipeline: Practical insights and 3D architectural perspectives
13mMLOps: From AI Research to Production-Ready Systems
42mMLOps 2025: A Journey into the Past and the Future
49mKubernetes and MLOps for Scalable and Reproducible Generative AI
13mMLOps at Scale
41mTHE STATE OF MLOPS - MACHINE LEARNING IN PRODUCTION AT ENTERPRISE SCALE