Reliable Machine Learning

Chapter 15. Case Studies: MLOps in Practice

This book has laid out principles and best practices for MLOps, and we’ve done our best to provide examples throughout. But there is nothing like hearing stories from folks working in the field to help see how these principles play out in the real world.

This chapter provides a set of case studies from different groups of practitioners, each detailing a specific issue, challenge, or crisis that they have lived through from an MLOps perspective. Each story was written by the practitioners themselves, so we can hear in their own words what they went through. We can see what they faced, how they dealt with it, what they learned, and what they might do differently next time. Indeed, it is striking to see how things as deceptively simple as load testing, or as seemingly unrelated as a launched update to an entirely different mobile app, can cause headaches for those in charge of daily care and feeding of ML models and systems. (Note that some of the details may have been glossed over or omitted to protect trade secrets.)

1. Accommodating Privacy and Data Retention Policies in ML Pipelines

By Riqiang Wang, Dialpad

Background

The automatic speech recognition (ASR) team at Dialpad is responsible for the end-to-end speech transcription system that generates live transcripts for various AI features (collectively known as Dialpad AI) for our customers across the world. Various subcomponents of our AI system heavily rely on the ASR outputs to ...