The research behind Meddies clinical AI.
About 99,000 synthetic Vietnamese hospital administrative scenarios, with 24 structured fields covering facility, situation, and constraint dimensions.
Read moreA Vietnamese dataset of 5,214 clinical-reasoning tasks across 8 hospital domains and 4 audiences, distilled from 444,694 candidates at a 1.17% acceptance rate.
Read moreA Vietnamese red-team set: 22,336 synthetic patient queries across five unsafe response modes, with 19,085 doctor-LLM responses passing an LLM-as-judge filter.
Read moreAn open research model for multilingual clinical de-identification. It separates patient identifiers from clinical data before the data moves through any AI processing.
Read moreA monolingual Vietnamese medical question-answer dataset in five domain groups: 2,941,561 paired QA rows and 7,109,244 question rows, CC-BY-NC research use only.
Read moreA synthetic dataset of 150,000 Vietnamese patient personas, used as context before any consultation or clinical note is generated. The fuller the patient context, the more realistic the synthetic output.
Read moreMedical AI scores near-perfect on exam-style benchmarks, but exam scores do not reflect real consultation quality. Meddies Consultant is a synthetic Vietnamese and English clinical-consultation dataset, built so models learn to ask rather than to conclude.
Read moreUse the dataset, run it through your pipeline, and tell us where it breaks. Or review the intended Meddies workflow.
Book a demo
