Software Misadventures Emmanuel Ameisen - On production ML at Stripe scale, leading 100+ ML projects, iterating fast, and much more - #11
6 snips
Jun 11, 2021 Emmanuel Ameisen, a machine learning engineer at Stripe and former lead at Insight Data Science, shares invaluable insights on building and deploying ML products at scale. He highlights common pitfalls in launching ML projects, emphasizing practicality over complexity. Emmanuel discusses the challenges of transitioning from research to engineering roles and the necessity of effective data management. He also touches on validating models in production, exploring testing methodologies, and shares his experience writing a book for engineers.
AI Snips
Chapters
Books
Transcript
Episode notes
Product Vision First
- Start ML projects with a clear product vision, not just a dataset or high accuracy.
- Define how your model adds value and what problem it solves before diving into complex modeling.
Patient ID Fiasco
- One fellow achieved 99% accuracy on patient outcome prediction.
- However, patient ID was a feature, leading to data leakage and a useless model.
Simplify and Iterate
- Instead of complex modeling, consider simpler solutions that achieve similar results.
- Iteratively refine the product and modeling approach to save time and effort.
