Stenobird

Episode

#356 The Forecast for Time Series Forecasts with Rami Krispin, Senior Manager of Data Science at Apple

Podcast
DataFramed
Published
Apr 20, 2026
Duration seconds
3213
Processing state
processed
Canonical source
https://www.datacamp.com/podcast
Audio
https://dts.podtrac.com/redirect.mp3/cohst.app/pdcst/6G1A6D/episodes.captivate.fm/episode/ac118883-e487-43e5-973f-06261907591c.mp3
JSON
/v1/public/podcasts/dataframed/episodes/356-the-forecast-for-time-series-forecasts-with-rami-krispin-senior-manager-of-data-science-at-apple
Markdown
/podcast/dataframed/356-the-forecast-for-time-series-forecasts-with-rami-krispin-senior-manager-of-data-science-at-apple.md

Actions

  • POST https://stenobird.com/v1/public/podcasts/dataframed/episodes/356-the-forecast-for-time-series-forecasts-with-rami-krispin-senior-manager-of-data-science-at-apple/transcription-requests
    Idempotently request low-priority transcript generation for this episode.
  • GET https://stenobird.com/podcast/dataframed/356-the-forecast-for-time-series-forecasts-with-rami-krispin-senior-manager-of-data-science-at-apple.md
    Read the agent-friendly Markdown representation of this episode resource.

Summary

Scaling time series forecasting requires moving beyond manual modeling toward automated, production-ready pipelines. This discussion explores the transition from traditional statistical methods to modern foundation models and the engineering rigor needed for deployment.

Topics

  • Time Series Forecasting
  • Foundation Models
  • Machine Learning Operations
  • Data Science Engineering
  • Predictive Analytics
  • Statistical Modeling
  • Production Pipelines
  • Model Monitoring

Highlights

  • Main idea: Foundation models for time series are emerging to solve the scalability crisis caused by massive, high-frequency datasets
  • Practical takeaway: Production forecasting requires robust monitoring, scheduling, and drift detection to ensure model stability over time
  • Failure mode: Relying on automated models without understanding underlying statistics can lead to catastrophic errors in sensitive business domains
  • Practical takeaway: Effective communication of forecast uncertainty and visualization is critical for building stakeholder trust
  • Main idea: The role of the data scientist is shifting from manual model building to designing complex, automated forecasting architectures

Chapters

  1. 1:00 The Rise of Time Series Foundation Models: An exploration of how new foundation models aim to assist data scientists in handling increasingly complex time series data.
  2. 5:00 The Scalability Challenge: Discussing the trade-offs between manual precision and the need to forecast hundreds of thousands of products simultaneously.
  3. 8:50 Handling Real-World Volatility: Why traditional models struggle with sudden structural shifts like policy changes or global events like COVID-19.
  4. 13:10 Risks of Automated Decision Making: The high stakes of overestimating demand and the dangers of delegating critical decisions to autonomous agents.
  5. 17:10 The Legacy of Prophet: A look at the impact of Meta's Prophet library and its role in introducing modern features to the community.
  6. 21:10 Backtesting and Model Stability: Using backtesting to ensure models remain reliable even when underlying data distributions change.
  7. 25:10 Engineering Production Pipelines: The necessity of building schedulers, data fetchers, and monitoring systems for live forecasting environments.