Solutions · Model labs
Training data that stands up to audit
Provenance is becoming a release gate. Mdexa keeps the license verdict, source risk and verbatim evidence for every medical dataset ready for your legal team.
A data inventory in, license and source risk per item out
Input
medqa-2 linglan https://huggingface.co/datasets/mirobody/ESL-Bench MIMIC-IV
batch_check · 3/4 matched
| Input | Card | License | Source risk |
|---|---|---|---|
medqa-2 | MedQA | Permissive | DerivedExam-derived |
linglan | LingLanMedBench Tradi… | Permissive | Exam-derived |
https://huggingface.co/datasets/m… | ESL-bench | Conflict | – |
MIMIC-IV | Not in the database yet | ||
Due diligence in one call, not page by page
Submit a list of names or URLs; each item comes back with its license verdict, source risk and card link. Items that cannot be matched with certainty are marked, never guessed.
Every verdict traces to source text and a verification date
Each piece of evidence records where it was found, the verbatim excerpt and the date. Excerpts are compared word for word with the fetched source snapshot.
Wired into agents and pipelines, so compliance moves into data preparation
Call it over MCP, or over HTTP with an API key you create in the console; every response carries license_status, last_verified and card_url. The license decision is made in the preparation script, not just before release.