Feature Store Startup Gets a VC Boost

Open source versions of feature stores are emerging that are designed help data scientists avoid reinventing the wheel. One feature store startup is attracting investors by adding a computing layer billed as helping to accelerate enterprise access to AI and advanced analytics.

Molecula Corp., the enterprise feature store startup, closed a $17.6 million funding round this week led by Drive Capital, with participation from new investor TTV Capital.

The three-year-old enterprise feature store developer based in Austin, Texas, said it has so far raised $23.6 million in two rounds. Early investors include Tensility Venture Partners.

Enterprise feature stores enable sharing and collaboration, helping harried data scientists to avoid duplicating the work of colleagues. The sharing of features stores is promoted as a way of making data science teams more productive.

“The feature store is emerging as the most transformative category in the data space because it automates the preparation of data for machine-scale analytics and AI,” said Molecula CEO Higinio Maycotte. Molecula promotes its approach as integrating data science tasks ranging from data preparation to MLOps.

Molecula’s platform extracts features, reduces data dimensionality and routes real-time feature changes into a central store. The startup claims that approach enables millisecond queries, computation and feature re-use without copying or moving raw data.

Molecula’s is the commercial version of the open source data format Pilosa, a distribured bitmap index designed to accelerate data analysis across multiple data sets. The startup’s distribution is designed to offer data scientists a cloud-agnostic data layer used for data analytics and AI applications.

“All other feature stores on the market today are built on reference architectures and are focused mostly on feature re-use,” the startup asserted in a blog post.

“This creates additional complexity, latency, speed and data gravity issues, and places limitations on how and where features can be used,” it added. The resulting platforms remain hard to implement, addressing some MLOps problems while “creating one more silo in an organization’s already-siloed architecture….”

Molecula’s approach differs, investors said, by creating a new computing layer for data that goes beyond simple data migration to the cloud. That approach provides a path to using AI and advanced analytics, said Andy Jenks, a partner with Drive Capital. “The market for data readiness is as big as it gets.”

Along with extracting and updating features, Molecula’s feature store is also touted as reducing data footprints by as much as 90 percent while providing secure data formats that promote collaboration.

Recent items:

Kaskada Accelerates ML Workflow with Its Feature Store

An Open Source Alternative to AWS SageMaker