Synthetic data for rare events

Pipeline

1. Profile real data and rare-event patterns
2. Generate synthetic records with statistical and LLM-based generators
3. Validate fidelity and privacy (no memorised records)
4. Train and test on real hold-out data only
5. Monitor drift once in production

Figures show the typical order of magnitude for this approach compared with calling a large general-purpose model. Actual results depend on the task and data; we measure them on your data during the baseline phase.