How to Machine Learn-ify any Product
summary
Replaces a ruley confirmation flow on Portal with a classifier that predicts which contact "John" means using organic confirmation labels from past prompts and features like ASR confidence NLU intent entity match and call/message frequency. Uses gradient boosted decision trees for low compute and nonlinearity. Offline plus A/B online evaluation and >95% precision objective drive a progressive rollout. Retrains on upstream model shifts user distribution changes or new features via active learning.
AI-written summary · about
chapters (7)
- 0:00Introduction to ML in products
- 2:52When to use machine learning
- 6:46ML model development cycle
- 12:28Evaluating model performance
- 16:38Deploying models to production
- 19:12Maintaining models over time
- 21:40Q&A session begins
related talks
15mMachine Learning for All
39mDeveloping Machine Learning for Impact
06mUnifying Online and Offline Worlds
40mMachine Learning helps to work with your test automation
37mAutomated Machine Learning: Building a Pipeline
01h 30mBecome a data-driven Organization with Machine Learning
30mDriving a Higher Level of Customer Support with Machine Learning
38mReal Life Customer Stories with AI