🤑 Paying Off Technical Debt in the Development of Machine Learning Apps
summary
Argument for incremental MLOps: a reproducible CI pipeline converts notebooks to scripts, pulls data from Snowflake, logs experiments to MLflow and stores data with DVC, then allocates SaaS GPU runners to train and publish model artifacts. A GitLab case study shows scheduled pipelines for drift detection, a model registry with promotion and canary serving, and faster onboarding via standardized container images. The approach reduces repeated manual retrains and makes experiments traceable and deployable in production.
AI-written summary · about
chapters (9)
- 0:00Introduction and motivation
- 3:34Identifying technical debt in ML
- 7:48Building a blueprint for ML workflows
- 11:21Scaling practices in larger teams
- 15:13Automating data scientist workflows
- 19:09Key use cases and integrations
- 24:31Collaboration and traceability benefits
- 27:43Summary and future directions
- 28:04Q&A session
related talks
24mFrom Model to Production: Automating the ML Lifecycle with CI/CD
23mAccelerating Model Deployment with MLOps
25mBuilding Scalable MLOps Pipelines That Actually Work
56mDevOps for Machine Learning
13mMLOps at Scale
59mDevOps for Machine Learning Projects
39mContinuous Delivery for Machine Learning — The Future of MLOps
20mBringing CI/CD Practices to Machine Learning with MLOps