A Declarative Approach to ML

Machine Learning (ML) is exploding in the enterprise as business users, engineers, and data practitioners increasingly look to leverage ML across all types of workflows. Yet, when deciding how to deploy ML, enterprise teams are faced with choosing the lesser of two evils. On the one hand, sophisticated teams can leverage open source and point solutions – however this is brittle, requires a ton of glue code, and has slow time to value. On the other hand, AutoML has emerged over the last several years to fill the gap with the promise of simplicity and fast time to value. In the customers we speak to, there is excitement around simplicity and low code interfaces, but far too often the models that get built with AutoML platforms get stuck as prototype deployments – as these platforms are blackbox and don’t have the robustness or fine grained control production users are looking for.

Enter Predibase, which is building the world’s first declarative ML platform to provide an alternative to AutoML – state-of-the-art machine learning for engineers and data practitioners as easy as writing a SQL query. Predibase’s declarative ML platform builds on top of an open source foundation – Ludwig and Horovod, two projects developed at Uber that together have 20,000 combined Github stars and over 100 open source contributors. These projects have been embraced by the OSS community and leading ML teams everywhere and power the simple yet powerful Predibase product experience.

Greylock is thrilled to lead Predibase’s Seed and Series A funding, and I’ve joined the Board.

Low-Code Machine Learning

With Predibase, users can build and deploy cutting-edge ML seamlessly, using its declarative end-to-end platform that goes from data to a deployed model. The basic idea with declarative systems is users specify model pipelines as configurations – allowing them to focus on the “what” and letting the system automate the “how.” Integrating with popular sources like Snowflake, Google BigQuery, S3, GCS, Delta Lake, and many more, Predibase users can train state-of-the-art models like BERT or GPT-2 on huge amounts of data through the platform or programmatically in just a few lines of code. Models can be deployed in one click and accessed via REST endpoints, through a Python SDK, or through PQL (a SQL-like syntax for machine learning).

Predibase has developed elegant abstractions and interfaces that nail what the market has been demanding – the simplicity of low-code/no-code and SQL, with no compromises on power and performance. The team initially developed the ideas behind declarative ML systems and applied them at Uber and Apple with great success in increasing productivity and performance. Since founding in 2021, Predibase’s declarative ML platform has already been deployed with several Fortune 500 companies with exciting results. With Predibase, companies are automating their ML use cases and seeing their production pipeline drop from months to days.

A Team of Distinguished Technical Talents

The Predibase co-founders – Piero Molino, Travis Addair, Devvret Rishi, and Chris Re – are a special team who we’ve been looking for ways to partner with for years. The team ties cutting edge ML expertise with incredible user empathy and design elegance. Piero (CEO) is the author of Ludwig and a long time architect of declarative ML systems, Travis (CTO) led the deep learning infrastructure team at Uber and was the lead for the Horovod project, and Dev (CPO) was the lead PM for Kaggle at Google. We are delighted to again partner with Chris (Co-Founder), an Associate Professor of Computer Science at Stanford University, co-lead of the DAWN group, and serial entrepreneur — co-founder of Lattice (acq. by Apple), Sambanova, and Greylock-backed Snorkel.

At Greylock, we have deep conviction around the power AI/ML can have on transforming the way humans work with machines. That’s why we have a history of backing founders building companies with ML/AI at the center, including Abnormal, Adept, Cresta, Inflection, and Snorkel, at the earliest of stages. We’re thrilled to support Predibase on their mission to bring high performance, low-code machine learning to users everywhere.