Tecton Feature Store

Commercial Information

Vendor Name

Tecton Feature Store

History

Founded by the creators of Uber's Michelangelo platform

Stand-alone vs. Platform

Stand-alone feature store, integrates with 3rd party MLOps platforms

Delivery Model

Fully-managed cloud service

Clouds Supported

AWS (now), GCP and Azure (roadmap)

Pricing Model

Consumption Pricing

Service Level Guarantees

Uptime, Serving latencies

Support

24 x 7 support & response time guarantees

Feature Store Capabilities

Feature Definitions

Declarative framework for defining features (incl. transformations and materialization)

Feature definitions are backed in git for central version control and CI/CD integration

Automated Transforms

Automated pipeline orchestration

Managed Batch, Streaming and Real-Time Transformations

Automated backfill of historical data

Pipeline visualization

Feature Ingestion

Spark/Pandas batch feature ingestion into offline & online store

Spark Streaming feature ingestion into offline & online store

Storage and Feature Processing Infrastructure

Online storage: DynamoDB

Offline storage: S3

Feature Processing: Spark and Python

Feature Sharing and Discovery

Web UI

Searchable feature catalog with metadata

Feature discovery including transformations, data lineage, and values

Feature versioning and dependency management

Training Dataset Generation

Dataset generated from offline storage using Python SDK

Row-level time travel

Online Serving

Serving endpoint / API for online data

Monitoring and Alerting

Managed data drift detection (roadmap)

Data quality monitoring (roadmap)

Monitoring of serving latencies and uptime

Security and Data Governance

Data remains in end-user's cloud account

ACL and RBAC (roadmap)

SSO

Data encryption at rest and in flight

Integrations

Batch data: S3, Hive/Glue, Redshift, Snowflake

Streaming data: Kafka, Kinesis

Tecton Feature Store FAQ

  • What is Tecton feature store?

    Tecton is a feature store founded by the creators of the Uber Michalangelo platform. It is a stand-alone platform that integrates with other 3rd party MLOps tools.

  • What is the difference between online and offline materialization in Tecton?

    Offline materialization corresponds to calculating and storing historical values of features and is used to speed up the computation of features when generating training data or doing batch inference. Online materialization involves persisting features to a fast storage solution (like Redis or DynamoDB), where they can be retrieved for real-time inference.

  • What kind of alerting does the Tecton Feature Store have?

    Tecton will send users alerts via email for common issues with feature pipelines. Some of these issues include failed jobs, high serving latency, or lower than expected feature freshness.

  • Who is the Tecton Feature Store for?

    Data scientists, data engineers, and machine learning engineers who need to serve complex/precomputed features to real time models, often with very recent freshness. For Tecton in particular, it is for those users looking to have a hosted and managed ready to use platform so developers can focus on building and deploying features and really accelerating the deployment of ML models rather than having to focus on ML infrastructure.

  • What data sources does the Tecton Feature Store work with

    s3/redshift/snowflake/kafka/kinesis

  • Is the Tecton Feature Store a managed version of Feast?

    No. The codebases are separate. However, Tecton fully believes in the Feast Feature Store project and is the main contributor.