Kumo Snowflake Native App Architecture White Paper

This document is a white paper detailing the architecture of Kumo, a Snowflake Native App designed for predictive analytics. Kumo enables users to specify learning tasks using a SQL-like predictive query language, automating data preparation, model training, and optimization processes. The architecture integrates large language models and graph neural networks to enhance predictive capabilities. Kumo operates within the Snowflake environment, ensuring that data remains secure and compliant with existing governance rules. The paper outlines how Kumo improves model performance by eliminating the need for manual feature engineering and allows for rapid generation of predictions. It describes the control plane and AI engine components that facilitate the execution of predictive queries and model building. The document emphasizes Kumo's ability to learn directly from relational data in Snowflake, providing users with advanced tools for analytics while maintaining full control over their data.