Product | Kaskada

Feature engineering & machine learning success

Feature engineering is the process of creating meaningful variables from raw data to power predictive machine learning models. High quality features are the single most important factor in a machine learning project’s success. Yet most data teams lack common tools and processes to create and deliver high quality features quickly and consistently.

Kaskada provides a unified platform for feature engineering, including a collaborative interface for data scientists and powerful data infrastructure to compute and serve features in production environments.

How It Works

Connect to Data

Ongoing access to all the data needed for feature creation

The Kaskada platform:

Ingests historical data from data warehouses and data lakes
Consumes messages from streaming data sources, like Apache Kafka and Amazon Kinesis
Transforms event-based data into a usable format for aggregations and other feature calculations

Design and Visualize Features

Create features using a simple, interactive interface

Data Scientists:

Write feature calculations using data from connected sources
Visualize feature distributions automatically and drill into outliers
Scale, transform, and encode features with a few clicks

Select and Export Features

Discover and share features to train models

Data Scientists:

View all features for a specific model in one central location
Understand how features are computed and share features across teams
Choose features to export for training and to promote to production

Deploy to Production

Deliver features directly to customer-facing applications

Data Engineers:

Call the feature store API to receive real-time feature vectors
Test and monitor feature performance over time from the Kaskada interface
Update the feature versions used in production in minutes