Predibase recently came out of stealth and this week we had the CEO and founder Piero Molino on the pod to talk to us about declarative ML and how espresso is the only type of coffee there is.

Who - Piero is one of the most accomplished professionals in ML systems. In stints at Uber ATG, Geometric Intelligence, and Stanford, he has made major contributions to the ML system OSS ecosystems through his papers on “Declarative ML”, and also open sourced the tool Ludwig before the MLOps Community was born.

Ludwig - Ludwig is a powerful interface for training models. it allows you to build models with a data type-driven approach that is extensible and interpretable. It is a toolbox that allows users to train and test deep learning models with minimal code.

Ludwig is an example of declarative machine learning. Declarative machine learning became popular after Piero's paper of the same name. It is an ML paradigm that focuses more on domain experts over model builders. It allows inexperienced users to easily train models, but also allows more advanced users the ability to tinker with the underlying structure of machine learning systems.

So where does Predibase fit into all this? It's an evolution of Ludwig. Predibase is opinionated and extensible. It’s following a new trend of companies like postgresML and Continual doing ML in the data warehouse. As the data warehouse becomes more central to business operations (both in terms of inputs and outputs), it also is becoming a popular destination to perform machine learning. This is because the compute costs are cheap and the outputs of the model can easily be integrated right where business logic and data exist.