Kawn Embed Medical

Kawn Embed Medical is a 139M-parameter Arabic text embedding model built for healthcare. It maps clinical notes, medical records, patient questions and health guidance into one vector space, so a query written in everyday Arabic finds the passage written in the precise language of professional medical text.

General-purpose embedding models meet medical Arabic as rare vocabulary. Kawn Embed Medical is built for it: clinical terms, diagnoses, procedures and findings, together with the everyday words patients use for the same things. The result is retrieval that holds up in a clinical setting, where a near match is not close enough.

It is the retrieval layer for Arabic healthcare search, for RAG over medical documentation, for sorting and classifying patient text, and for finding records that repeat each other. The weights are open, and the model runs behind the Misraj cloud API or inside your own environment.

• Semantic search across clinical notes, discharge summaries, guidelines and patient records, in Arabic.
• Retrieval for RAG: the layer that grounds a medical assistant in your own documentation.
• Everyday to clinical: a question in colloquial Arabic matches the passage written in professional medical Arabic.
• Classification and clustering of patient messages, referrals and reports by topic.
• Duplicate detection across records that say the same thing in different words.