Fast and Reproducible: Taming AI/ML Dependencies

This talk addresses the critical but often underrated aspect of AI/ML systems: dependency management. Savin Goyal explores common pitfalls that lead to instability in production deployments and slow down development speed. He shares learnings from working with organizations like Netflix and Amazon, demonstrating how to create fast, stable, and reproducible environments for experimentation. The discussion covers scaling dependencies in cloud environments, handling non-Pythonic dependencies, auditing the software supply chain, and accelerating prototyping with fast container image building.