Tecton: Real-time feature stores connecting data to ML models

An Intellyx Brain Candy Update

AI is perhaps the fastest moving technology space we cover right now, and Tecton has been off to the races since our last coverage of them in 2020 building a highly performant real-time feature store service that strings machine learning models between multiple data sources and applications.

If you aren’t familiar with the concept of a real-time feature store, consider an example of a data scientist in a bank attempting to detect and prevent fraudulent transactions. Rather than custom extracting data from Kafka streams and warehouses, and then coding their own inference subroutines for cloud implementation, the practitioner could have Tecton bring together 10 data sources and several current ML models, and insert a fast fraud checkpoint from the feature store.

This modular reusability can be useful for analytics workflows, to be sure, but the most compelling uses of these features are in operational scenarios like customer price quoting or an insurance claims settlement where the tolerance for latency and current information in making decisions is at a minimum.

Also of note was how ML features can be delivered as part of a broader DevOps style collaborative pipeline, so multiple teams can share and continuously redeploy what works, while having changing conditions and model maintenance handled by the feature store.

 

©2021 Intellyx, LLC. At the time of writing, Tecton is not an Intellyx customer. Want to see more BrainCandy? Subscribe today. If you are a vendor seeking coverage from Intellyx, please contact us at PR@intellyx.com.