The evolution of MLOps at Cian
summary
The talk describes how Cian scaled MLOps for over 60 ML models by moving from ad-hoc manual processes to an automated ML platform. It covers their use of automated pipelines for code changes, model deployment, and scheduled retraining, explains why they built a custom ML Model Server for online inference, and how managing model workflows through Jira simplified and sped up model development and delivery.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 3:56Early stage problems
- 10:00Automation and unification
- 13:14Template standardization
- 24:24Infrastructure selection
- 27:36Model build and deploy
- 31:57Pipeline management
- 38:17Results and conclusions
- 42:55Q&A
related talks
10mFrom Model to Production: Building Scalable MLOps Pipelines
53mExperience with an end-to-end MLOps pipeline: Practical insights and 3D architectural perspectives
05mThe Evolution of MLOps
20mBringing CI/CD Practices to Machine Learning with MLOps
41mMLOps in 30 Minutes
25mBuilding Scalable MLOps Pipelines That Actually Work
43mNavigating the MLOps Journey: Key Considerations for Successful Implementations
13mMLOps at Scale