DataFramed #356 The Forecast for Time Series Forecasts with Rami Krispin, Senior Manager of Data Science at Apple
24 snips
Apr 20, 2026 
Rami Krispin, Senior manager of data science and engineering at Apple who builds production forecasting systems and teaches time series, joins to discuss scaling forecasts across many products. He covers foundation models for time series, when to trust automation versus manual feature engineering, backtesting and model selection, production pipelines and monitoring, and how to communicate forecast uncertainty to stakeholders.
AI Snips
Chapters
Books
Transcript
Episode notes
Foundation Models Solve Forecasting At Scale
- Time series foundation models primarily solve scale by learning from many datasets so teams can forecast thousands of series with less manual effort.
- Rami contrasts old TS tooling (R's ts objects) with foundation models trained across diverse use cases to handle massive modern data volumes.
Walmart Example Shows Why Scale Matters
- Rami recounts retail examples like Walmart needing forecasts across thousands of SKUs for capacity and inventory planning.
- He warns scaling trades some accuracy for broad visibility but yields major cost savings in storage and spoilage.
Feature Engineering Is The Business Edge
- Feature engineering is the critical differentiator in business forecasting when historical patterns break.
- Rami notes regressors and engineered event features matter most when policy, product launches, or shocks (like COVID) disrupt seasonality.
