Delivery

Delivery

Purpose: expose predictions to end-users with low-latency and generic, reusable semantics How: Pyprimed + client-side libraries

Models

Dense models

One-to-many models

Sparse models

Campaigns

A/B testing

A/B membership assignment

WRANDOM

STICKY

FORCED

Control

Random

Null

Custom A/B variants

Blending

Recency

Dithering

Conversion

Publishing

Predictions

Upsert

Delsert

Targets

Upsert

Delsert (???)

Advanced

Authentication