← All components
Machine learning
ML / recommendations
Trains and serves models for ranking, recommendations, ETA and fraud.
When to use
- Personalised feeds
- Pricing, ETA prediction, risk scoring
Pitfalls
- Feature skew between training and serving
Options by provider
The first option is what diagrams show for that provider.
Amazon Web Services
- Amazon SageMakerManaged
- Amazon PersonalizeManaged
Google Cloud
- Vertex AIManaged
Microsoft Azure
- Azure Machine LearningManaged
Open source / self-hosted
- KubeflowSelf-hosted
- MLflowSelf-hosted
- Feast — feature storeSelf-hosted
Used in
- YouTube
- Recommendations — Ranks the home feed and related videos from watch history
- Netflix
- Recommendations — Builds personalised rows for each profile
- Amazon
- Recommendations — 'Customers also bought' and personalised picks
- Payment gateway
- Risk engine — Scores each payment for fraud; may block or require 3-D Secure
- Google Maps
- ETA model — Predicts travel time from live and historical speeds
- Airbnb
- Ranking & pricing models — Orders results; suggests prices to hosts
- NoBroker
- Verification & fraud — Flags brokers, duplicates and fake listings
- Zomato / food delivery
- ETA & prep-time models — Predict cooking and travel time
- Quora
- Ranking models — Answer quality and feed relevance
- LinkedIn
- Recommendations — People you may know, job matches, feed ranking
- Buy now, pay later
- Risk & fraud models — Scores users and transactions