Accelerating ML impact
summary
Transforms an underperforming RBC ML team into a high-output engine by pushing 60 models in production in 6 months, cutting end-to-end model delivery from months to weeks, and reporting 30–45% model improvement over previous baselines. Achieves 8x compute reduction and 10x storage reduction, replaces a 500-feature random forest with a 20-feature logistic model for ~40% better performance and far greater stability, and emphasizes stabilizing pipelines, parametric models, and business-level evaluation metrics.
AI-written summary · about
chapters (13)
- 0:00Introduction and background
- 1:23Achievements and impact at RBC
- 2:26Best practices and pitfalls
- 3:49Avoid waiting for perfect solutions
- 5:15Focus on integration points
- 7:03Beware of bleeding-edge hype
- 8:45Build stable foundations first
- 9:25Avoid trial-and-error mentality
- 10:51Principles over trends
- 12:37Power of parametric models
- 15:09Principle of parsimony
- 16:51Optimize for real impact
- 20:03Know your data and limits
related talks
32mMaking Models Work: How an optimized ML lifecycle can drive better ROI
26mManaging ML Models @ Scale
37mAutomated Machine Learning: Building a Pipeline
41mChallenges in incorporating ML models
38mFantastic ML apps and how to build them
25mBuilding Scalable MLOps Pipelines That Actually Work
32mTeam Teardown: Evolving an ML Platform
10mFrom Model to Production: Building Scalable MLOps Pipelines