FactSet says serverless GPUs cut custom model workflow from days to hours

FactSet reduced the time to set up, train and deploy a custom Text-to-Formula model from days to hours using Databricks AI Runtime, according to a Databricks case study quoting a FactSet engineer. The system translates investment professionals' requests into FactSet Query Language formulas. Engineers select serverless GPU compute for each training run without provisioning the underlying environment, while keeping data governance and model tracking within existing Databricks workflows. FactSet fine-tunes models roughly every two to three weeks. The reported time reduction covers the combined setup, training and deployment workflow.