SDV (Synthetic Data Vault) is a platform for synthetic data generation, first developed at MIT and now the core product from DataCebo. The Synthetic Data Vault is delivered as a python library and runs on any environment which can run python, even fully locally with no data leaving the environment. There are two different versions of The Synthetic Data Vault each built with different use cases in mind:
SDV Community - Designed for academic purposes and to help people get started with synthetic data, the community version is perfect for experimentation and smaller projects. Best suited for up to 5 tables with the depth of 2.
SDV Enterprise - A powerhouse in the world of synthetic data, its designed for your most complex enterprise level tasks. SDV Enterprise handles complex multi-table schemas with ease, with no limit on the depth or table count.
At their core both SDV Enterprise and SDV Community follow a simple three step framework, regardless of the model you pick.
The Learning Phase (Fit), you pass your real data (single or multi table) to a synthesizer that runs entirely on your infrastructure. The synthesizer scans the data to understand correlations, distributions and constraints. Your raw data never touches any external servers, it stays exactly where it is.
The Generation Phase (Sample), once trained, the model creates an entirely new dataset. Fully synthetic copies, entirely new rows generated from scratch that follow the same rules and patterns as your original data. This phase can also run fully on-prem as well.
The Quality Check Phase (Evaluate) SDV comes with built-in metrics to compare the synthetic data against the real, ensuring both privacy and fidelity. All within your environment.
Basically SDV allows you to train your own generative model for tabular data and use it as much as you would like, even move the model across environments or share it with others. If you are interested in more information on how SDV works you can find out more at: docs.sdv.dev