Flock Safety, a company specializing in AI-driven public safety solutions, wanted better model reproducibility and explainability of their machine learning models. Flock’s models analyze data from network-connected cameras to help solve crimes. The data for these models can quickly grow, and maintaining consistent training datasets and understanding the impact of data changes was becoming increasingly complex for Flock. The lack of robust version control for structured data in traditional ML tools hindered their ability to reproduce models accurately and collaborate efficiently among data science teams. To address these issues, Flock adopted Dolt, a version-controlled SQL database, as their feature store. Dolt's Git-like capabilities allowed them to tag specific versions of their training data, ensuring that models could be reproduced with the exact datasets used during training. The ability to compute data differences between versions down to the cell level enhanced model explainability by highlighting how changes in data influenced model outcomes. Flock also uses Dolt's branching and merging features to create isolated development environments for different teams, enabling parallel workflows without conflicts. As new data comes into the main branch, it can be synced as needed to the development branches using Dolt's merge functionality. This same merge functionality allows successful projects' data to be merged into the main branch when it's ready.