Streaming aggregation - the spark behind real time ML

Real time ML tasks, such as credit card fraud detection, recommendation systems, anomaly detection and others use features that are computed as aggregates over real-time data streams.Data freshness and low latency are key in these use cases, yet achieving them often results in a resource-intensive solution and an angry CFO.Join Gal Lushi and Yoni Ben-Dayan as they take us through how Qwak’s engineering teams built a turn-key streaming aggregation solution that addresses challenges, while guaranteeing:

  • High data freshness, low latency and high throughput
  • EXACTLY ONCE and support for late arrivals.
  • Efficient handling of multiple small and long time windows.
  • Consistency between inference and training (because who wants a training-serving skew?)
Gal Lushi

Gal Lushi

Tech Lead, Feature Store

Yoni Bendayan

Yoni Bendayan

Software engineer, Feature Store

Watch Now

Qwak optimizes ML Model Production for
ML driven organizations.

Here is what our customers have to say about us:

"From the get go, it was clear that Qwak understand our needs and requirements. The simplicity of the implementation was impressive.
Automatic deployment and continuous training were crucial to allow us to scale. Qwak gave us a type of "Jenkins" for machine learning."

Shaked Zychlinksi Head of Recommendations Research

"Using Qwak allowed us to focus on creating a business impact rather than spending valuable time on our infrastructure setup.

At JLL our development is very time sensitive. As a result of implementing Qwak, we improved our execution time by 4.5X."

or hiltch
Orr Hiltch, Vice President of Engineering

"With Qwak we were able to improve our ML delivery dramatically.

Qwak has allowed us to work to the highest engineering standards from day one and to invest the majority of our efforts in our business challenges and not into plumbing."

Elad Silvas Data Science Manager