This document describes some sample use cases where users have employed Rockset.
A single data scientist built this end-to-end in a couple of days, resulting in a solution that is live, extensible, and easy to integrate with new data sets in the future. Projects of similar scope previously took a team of 2-3 about a month to build without Rockset and gave rise to a one-time static report, which was not repeatable or easily maintainable.
Real-time Adobe Omniture data is continuously ingested into Rockset, with custom data retention of 30 days. Rockset automatically purges older data. User and Product dimension tables are continuously live synced from operational databases, and only the latest copy is kept. Various personalization features (e.g. previously visited products, product characteristics of interest to individual users, etc.) are extracted via live millisecond SQL JOINs between clickstream and User/Product tables. These real-time personalization features are fed to the app and app’s relevant ML models to personalize the user experience.
Building new personalization features is simple and fast, powered by a single integrated data management system. Personalization systems working on User and Product data are fully isolated from the e-commerce site’s operational databases, thereby allowing for more rapid experimentation and A/B testing.
Rockset’s live write API is used to stream real-time connected device JSON data, with a custom retention of 90 days. A data forwarder was built to sync JSON API data from the ticketing system with Rockset. User data was continuously live synced from Postgres. A simple Python Flask internal application was built (few hundred lines of Python). The app takes in a customer ID and executes fast live search, aggregation, and JOIN queries across all the data sets to display a single-pane view to improve customer support.
With a fast unified customer support portal, time to diagnosis and resolution for support tickets improved significantly. The organization previously had many of these data sets in a traditional warehouse, which could not handle real-time data and was too slow to power interactive search and aggregations. Due to these challenges, customer support involved long wait times for page loads and many unresolved support issues.