Generate
Create synthetic datasets across tabular, text, image, and time-series data from a sample, a schema, or a scenario description.
Platform
The synthetic data platform for ML teams.
Generate, control, evaluate, and deliver synthetic training data from one workspace — built to fit the way ML teams already work.
Control
Steer what gets generated: rebalance classes, condition on attributes, set difficulty, and dial up the edge cases.
Evaluate
Check synthetic output against reference data for similarity and usefulness before it ever reaches a training run.
Deliver
Export to the formats and storage your pipeline already uses, and version datasets alongside your experiments.
Control
Design the dataset, don't just sample it.
Real data reflects what happened. Training data should reflect what your model needs to learn. SETC lets you specify the mix.
- Rebalance target classes and under-represented segments
- Condition generation on attributes, domains, or scenarios
- Increase the share of hard and rare examples
class distributionafter rebalancing
Source
Synthetic
Evaluation
Know it's useful before you train on it.
Every dataset comes with a report comparing it to reference data, so you can judge fit for your task instead of trusting it blindly.
- Distribution and correlation comparisons
- Train-on-synthetic, test-on-real style checks
- Privacy checks for closeness to source records
evaluation_reportsummary
Column distributions✓ pass
Pairwise correlations✓ pass
Downstream task check✓ pass
Nearest-record distance✓ pass
Developers
Fits into code, not just a dashboard.
Generate datasets from notebooks, scripts, or pipelines. Treat synthetic data as another versioned artifact in your ML workflow.
generate.py
from setc import Client client = Client() dataset = client.generate( source="s3://ml-data/support_tickets.parquet", rows=250_000, rebalance={"label": "uniform"}, ) dataset.evaluate() # compare against source dataset.export("parquet")
Get started
See SETC on your own use case.
Bring a dataset or a problem statement. We'll walk through how SETC would generate, control, and evaluate data for it.