Dashboard
AI-assisted mapping of raw clinical source data to SDTM-style target standards, with human approval, full traceability and deterministic validation.
Workflow
- Sources: profile the raw datasets (types, sample values, missingness).
- Mapping: generate AI mapping suggestions per source variable, then a person accepts, edits or rejects every single one.
- Spec: approved decisions form a traceable mapping specification (exportable as CSV / JSON evidence package).
- Code: readable transformation code is generated from the spec and can be executed against the sample rows.
- Validation: deterministic conformance checks (required variables, controlled terminology, duplicates, ISO dates) run against the standardized output.
Recent activity
Source metadata browser
Live profiling of the raw source datasets: inferred type, sample values, distinct count and missingness are computed from the synthetic sample rows.
Variables
Sample records
raw synthetic rowsMapping workspace
Generate AI mapping suggestions per source variable, then accept, edit or reject each one. Nothing becomes part of the specification until a person decides.
Mapping specification
Every approved mapping with its transformation logic, controlled terminology and the person who approved it. Exportable for downstream review.
Approved mappings
Generated transformation code
Readable pandas-style Python generated from the approved mapping specification. “Execute transformation” applies the same logic in the browser to the sample rows.
Validation
Deterministic conformance checks over the standardized output: required variables, controlled terminology, duplicates, subject identifiers and ISO 8601 dates.
Validation run history
History & traceability
Complete audit trail plus the AI execution log: every model call recorded with prompt version, input hash, confidence and latency for reproducibility.