Databricks releases open-source real-time next-best-offer recommender accelerator
Start here
Databricks published an open-source Solution Accelerator that builds a real-time next-best-offer recommender for retail banking using Feature Views, defining features once for both training and serving. The accelerator combines batch features with streaming in-session intent from Kafka, trains a LightGBM ranker with point-in-time-correct joins, and serves all 40 offers through a route-optimized Model Serving endpoint. In an untuned smallest-configuration benchmark, Databricks reports that streaming events became visible to the model in roughly a tenth of a second. Everything runs on serverless compute governed by Unity Catalog; the streaming portion additionally requires a Unity Catalog Kafka connection and Amazon MSK connectivity. Ranking endpoints use scale-to-zero, so the first request after an idle period can incur a cold start of tens of seconds.
Generated from reports · updated 1 h ago
Report timeline
Follow the reports to see the event from different sides.
- Databricks BlogBuilding a Real-Time Next-Best-Offer Recommender with Databricks Feature Views
Databricks published an open-source Solution Accelerator that builds a real-time next-best-offer recommender for retail banking using Feature Views, defining features once for both training and serving. The accelerator combines batch features with streaming in-session intent from Kafka, trains a LightGBM ranker with point-in-time-correct joins, and serves all 40 offers via a route-optimized Model Serving endpoint. In an untuned smallest-configuration benchmark, streaming events became model-visible in roughly a tenth of a second.
Heat for this event
Not enough continuous observations yet to draw a trend.