How Preply Reduced ML Model Development Time from One Month to One Day
summary
A Preply senior engineer explains how a small team inherited an existing ranking model and reworked their ML development workflow so new model iterations now take one day instead of one month. The talk describes concrete technologies and processes the team introduced in their Data Science practice to dramatically speed up model development.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:47Initial ML dev state
- 3:35Data and dev issues
- 5:22SQL and ETL optimization
- 6:47SageMaker drawbacks
- 8:58Databricks benefits
- 11:03Pandas to PySpark
- 13:09Model code issues
- 15:58Custom Feature Store
- 19:09MLOps framework
- 21:41MLOps in action
- 24:10Limitations and future
- 26:00Results: dev time
- 27:02Q&A
related talks
22mAccelerating ML impact
32mTeam Teardown: Evolving an ML Platform
26mSmall teams, big dreams: How you can get started with AI today
28mAccelerating AI at Scale with Data
57mGood Practices for Applied Machine Learning
32mMaking Models Work: How an optimized ML lifecycle can drive better ROI
28mBuild an End-to-End ML Workflow for Your Organization
27mML Platform Journey