Sign In

Generation Products

RadMah exposes distinct products for distinct data problems. A route, SDK call, or approved ADS plan selects the product explicitly; no hidden model is allowed to impersonate another product.

Product map

ProductInput authorityCustomer outputUse it for
FabricateA natural-language data requirementReviewed schema and generated customer dataNew data when no source dataset exists
SynthesisAn owned CSV or Parquet dataset with 1,000+ rowsLearned synthetic data with quality and privacy evidenceStatistically faithful data derived from an existing dataset
FHIR R4 SynthesisAn explicit healthcare standards workflowFHIR R4 bundle artifacts with terminology governanceHealthcare interoperability and standards-conformant bundles
Virtual SCADAA plant/process simulation specificationPhysics-governed OT telemetry and protocol artifactsIndustrial digital twins, training, and control-system testing
ICS SecurityAn authorized attack or cyber-range scenarioGround-truth security telemetry and detection artifactsSOC, IDS, incident-response, and cyber-range validation

ADS plans; it does not replace the products

The Agentic Data Scientist may build a governed multi-step plan across these product surfaces. The exact plan and maximum credit authority are shown before execution. ADS cannot substitute a private engine or bypass the selected product contract.

Fabricate

Fabricate begins with a human requirement rather than a source dataset. The customer reviews and refines the generated contract before approving final generation. Fabricate is not the learned Synthesis runtime and it must not be used as an invisible repair or vocabulary layer inside Synthesis.

Open the Fabricate and Mock Data guide →

Synthesis

Synthesis learns a generic mixed-type representation from an owned source dataset and generates new rows from that fitted state. Its production path has one learned engine. Unsupported geometry fails explicitly instead of falling back to a different statistical model.

Public operating modes

ModePurposeTraining controls
QualityHighest supported learned qualityCustomer-authorized ceiling and advanced controls; held-out early stopping
AutomaticGoverned defaultsServer-owned controls and held-out convergence
FastBounded lower latencyServer-owned bounded ceiling with the same evidence authority

These modes are not model selectors

Every mode runs the same learned Synthesis engine. Selection cannot depend on industry, dataset identity, column names, values, tenant, benchmark identity, or a hidden vocabulary.
Run Synthesis end to end →

FHIR R4 Synthesis

FHIR R4 is a separately governed standards package. Its structural and terminology rules are explicit at the FHIR boundary and are not loaded by generic Synthesis. It is not a fallback model or an industry template inside the learned engine.

Virtual SCADA and ICS Security

Virtual SCADA produces physics-governed industrial process telemetry and protocol artifacts. ICS Security adds authorized attack scenarios, ground-truth security labels, and detection outputs. They are simulation products, not alternate paths for tabular Synthesis.