Rockfish generates fully synthetic data, either from a schema you describe or a model learned from your existing data: structurally sound, statistically faithful, and safe to share on day one. It also builds test environments for AI agents, with ground truth and automatic scoring. Each report below walks one example end to end.
Environments with ground truth, for testing and training AI agents.
Data generated purely from a schema. No source data needed.
A microservice call graph becomes a complete, correlated OpenTelemetry signal set: distributed traces, RED metrics, and logs, with a planted incident an SRE can find, and it answers the on-call questions a practitioner actually runs.
The full card-payment lifecycle across eight related tables: 10M transactions with every foreign key resolved, causal ordering, and the distributions a fraud team expects, and no real cardholder data.
A model trained on real source data, then used to synthesize new records.
Point Rockfish at a real electronic health record, train a model on it, and generate a synthetic replica across five linked tables with every foreign key resolved and no real patient, statistically faithful and ready to train real models on.
Point Rockfish at a live MySQL database, train a model on it, and generate a synthetic version that keeps the relationships across tables intact and carries no real personal data.