Espresso AI builds a neural compute optimizer that applies machine learning to performance engineering for data warehouses. Its platform uses ML models that read and understand code in order to rewrite and optimise individual SQL queries on the frontend, while allocating physical compute on the backend. The target is spend on cloud data platforms: the company says customers save up to 70% on Snowflake and Databricks bills, without changes to existing workflows.
Installation is designed to be minimal. Customers set up by running a single SQL command and adjusting a handful of configuration settings, a process the company puts at roughly ten minutes. Because the optimiser operates on queries and compute allocation rather than on user tooling, engineers do not need to alter pipelines or move workloads.
The technical work spans several fields: large language models, AI code understanding, neural scheduling and optimisation, SQL query optimisation and compute resource allocation. Espresso AI runs a research programme aimed at what it describes as superhuman code optimisation, framing the effort as a shift beyond conventional optimisation tools and the performance gains available under post-Moore's law constraints.
The firm is a small team, with members recruited from Google, Apple and MIT, and describes its mission as automating performance engineering in the manner of a round-the-clock team of data engineers. Founding year, funding and geographic footprint are not disclosed.
