A proprietary AI model trained on 5.5 million real clinical texts and augmented with 6.5 million validated medical terms from SNOMED CT, LOINC, and ICD-10. It runs within your Azure environment: your patients' data never leaves your organization.
Patient with a VP shunt. Follow-up lab tests scheduled in 7 days and review at the Neurosurgery clinic.
Patient with a ventriculoperitoneal shunt. Follow-up blood tests in 7 days and review at the Neurosurgery clinic.
Our model runs entirely within your Databricks environment on Azure, with no internet egress and no third-party AI services. Clinical accuracy with the data sovereignty your patients deserve.
5.5M
real clinical sentence pairs used to train the model.
6.5M+
validated medical terms from SNOMED CT, LOINC, ICD-10-ES, DEMCAT, and UMLS.
12+
languages available, expandable without retraining the model.
97%
terminology accuracy in every translation.
Inference runs on Databricks with Private Link: no internet access, no third-party AI, and no data residency risk.
5.5M sentence pairs from regulatory documents (EMEA), clinical corpora (MeSpEn), and biomedical literature (WMT). This isn't a general-purpose model adapted with prompts.
SNOMED CT (Spanish edition), LOINC, CIE-10-ES, DEMCAT, and UMLS, retrieved in real time and injected into the model's context.
Accuracy validated on WMT Biomedical, the international standard for evaluating medical translation.
* COMET (Crosslingual Optimized Metric for Evaluation of Translation) is a machine learning–based metric that compares the model's translation with human reference translations. It correlates better with the judgment of professional translators than classic metrics such as BLEU. It ranges from 0 to 1. Results measured on the WMT Biomedical test set, ES↔EN pairs, with 23,410 sentence pairs evaluated.
Drag in a PDF, DOC, DOCX, or TXT file, or paste the text directly.
Select the target language from the 12 available.
In 30–90 seconds: it detects medical entities (NER), retrieves terminology in real time, and translates using it.
Adjust whatever you need with track changes and download the result as a PDF.
See how it works in just a few minutes
Where it's processed
Third-party servers
Within your Azure tenant, with no internet access
Model
General-purpose, adapted with prompts
Proprietary: QDoRA on X-ALMA 13B with 5.5M clinical pairs
Terminology
Whatever the model happens to remember
Real-time RAG with over 6.5M validated terms
Clinical abbreviations
Literal translation
Expanded: DVP → ventriculoperitoneal shunt
Formats
Copy-and-paste text
PDF, DOC, DOCX, and TXT with structure intact
Review
You make corrections outside the tool
Editor with tracked changes before delivery
Integration and analytics
Manual
REST API for HIS/EHR and a usage metrics dashboard
Discharge summaries for international patients, medical records for international referrals, and multilingual informed consent forms.
Communication with medical tourism patients, second-opinion reports, and documentation for foreign insurers.
Results for foreign facilities and pathology reports with terminology kept intact.
How accurate is it with medical terminology?
It achieves a COMET score of 0.88–0.90 on WMT Biomedical, the international benchmark for evaluating medical translation. It's trained on 5.5M real medical sentence pairs and augmented with more than 6.5M validated terms from SNOMED CT, LOINC, and ICD-10, ensuring consistent clinical terminology across every translation. It also correctly expands clinical abbreviations (e.g., VPS → ventriculoperitoneal shunt).
Is patient data sent to external services?
No. All inference runs within your Azure environment through Databricks with Private Link, with no internet egress from the processing clusters. Your documents are never sent to OpenAI, Google Translate, DeepL, or any other third-party AI service. Data is processed only during translation and is never stored permanently.
Can it integrate with our HIS/EHR?
Yes. The platform provides a REST API with API key authentication for secure integration with hospital information systems and electronic health records. All API traffic stays within your private network.
How many languages does it support?
Currently 12: Spanish, English, French, German, Italian, Portuguese, Arabic, Chinese, Japanese, Korean, Russian, and Catalan. The base model is trained on many more languages, so we can enable others on demand without retraining it.
What document formats does it support?
PDF, DOC, DOCX, and TXT, with automatic text extraction and export to professional PDFs with customizable templates.
How does the translation process work?
Upload the document or paste the text and choose the target language. In 30–90 seconds, the model detects medical entities (NER), retrieves terminology in real time, and translates using those terms. Then you review, edit if needed, and download the result.
Does it include staff training?
Yes. All plans include an initial training session and complete documentation. No technical knowledge is required to use it.
Demo with your own documents
Deployment within your Azure environment
Initial training included