Compile all the research on BioInformatiCo’s Leova Synthetic Data Generator — runner-up in the CDISC AI Innovation Challenge — which digitises a protocol into USDM, generates study specifications and synthetic test scenarios, robotically enters them into an EDC (Viedoc in the demonstration), and checks the outputs with CDISC CORE; the company plans a hybrid free-plus-paid model.
This dossier curates 33 research items from 18 independent hosts — 6 on recognition and the challenge, 8 on the company, product and commercialization, and 19 on technical context and validation — selected from roughly 4,865 deduplicated search results across 28 queries. It is a broad public-web dossier through 1 November 2026, not every internal or non-public artifact; the most-cited host backs only 12.1% of items, and every row is single-URL sourced in the ledger below.
Each box is one stage of the Leova loop. Green stages are demonstrated or directly stated in the challenge demonstration; amber stages rest on company claims; dashed notes are unresolved due-diligence questions.
Public evidence supports a demonstrated platform or prototype under active development with a genuinely distinctive closed-loop testing approach — judges reportedly emphasised its standards integration and traceability. The evidence does not establish broad general availability, production-scale validation, or regulatory qualification: the Leova site is gated, the company itself calls it “the platform we are building”, and its own pages describe proprietary internal tooling used in service delivery, with “no software to license, learn or log into” (bioinformatico.com). Shirley Collie and Robert Kieser were scheduled to present at the CDISC US Interchange in Denver on 5 October 2026 (it-online.co.za). Commercialization is unresolved and should be confirmed with BioInformatiCo directly.
Leova converts a clinical trial protocol into a structured USDM representation, generates study specifications — forms, fields, edit checks, SDTM mappings — then builds repeatable synthetic participant scenarios including deliberately nonconforming records, enters them robotically into the actual EDC (Viedoc in the demonstration), and runs CDISC CORE conformance checks on the downstream datasets (it-online.co.za). Viedoc is the EDC used in the demonstration; no reviewed source names it a formal partner or customer.
The value is pre–first-patient: study databases, edit checks, listings, reports, monitoring workflows and transformations can all be tested before real participant data exists, and retested consistently as requirements evolve — the same need Takeda addressed with an LLM-enhanced test-data generator for Veeva CDMS (lexjansen.com).
Standards mapping: USDM structures study design in machine-readable form (kitsa.ai); CDASH informs data collection; SDTM structures submission tabulations; ADaM supports analysis datasets traceable back to SDTM (intuitionlabs.ai); CORE executes CDISC’s open conformance rules (cdisc.org).
Limits: CORE checks standards conformance only — it does not prove the protocol was interpreted correctly, that study-build logic is clinically correct, or that regulators have qualified Leova. “End-to-end traceability” remains a product design claim until independently demonstrated.
Comparable approaches already exist and frame what Leova must beat: the SYNDATA modular framework generates eCRF-conformant synthetic datasets for operational validation (mdpi.com); PharmaSUG 2026 work closes the loop on AI-generated SDTM mappings with CORE and synthetic data (pharmasug.org); LLM-plus-knowledge-graph pipelines produce synthetic ADaM datasets (lexjansen.com); and GPT-4o-mini with Pydantic generates structured synthetic trial data (pharmasug.org). None of this research establishes that any synthesis approach is superior, statistically equivalent to real data, privacy-protective, or regulator-qualified — those limits apply to Leova too.
Where each item comes from, with its section and whether the source itself flags a caveat. Below, the spread across the 18 hosts.
| # | Section | Topic | Date | Caveat | Source |
|---|
Method: curated public-web dossier on BioInformatiCo’s Leova Synthetic Data Generator — 33 research items across 3 sections (recognition 6, company/commercialization 8, technical context 19) from 18 hosts, selected from ~4,865 deduplicated results across 28 queries; cutoff 1 November 2026. Counts measure distinct curated items, not article impressions; the largest host backs 12.1% of items. Long topic names are truncated in the ledger for space; full findings text was cut. Not an audit of internal or non-public validation artifacts.