The challenge
The clinic had wired ChatGPT (GPT-4) into a Whisper transcription pipeline as a DIY scribe. It hallucinated medication names and dosages on roughly 1 in 8 notes.
Norwegian dictation was constantly mistranslated into Swedish or English medical terminology — every note required a full clinician rewrite before it could be signed.
Before S10.AI
- The clinic had wired ChatGPT (GPT-4) into a Whisper transcription pipeline as a DIY scribe. It hallucinated medication names and dosages on roughly 1 in 8 notes.
- Norwegian dictation was constantly mistranslated into Swedish or English medical terminology — every note required a full clinician rewrite before it could be signed.
- No BAA, no data-processing agreement, no EU data residency — the setup was quietly out of compliance with GDPR and the Norwegian Personopplysningsloven.
After S10.AI
- CRUSH deployed in five clinic days with native Norwegian bokmål, Swedish and Danish medical vocabularies — no translation layer between speech and note.
- Automatic language detection per encounter: the clinician speaks naturally and the finished note lands in the correct language and the correct ICD-10-NO / KSH-97-P taxonomy.
- Structured write-back into System X via clinical RPA — SOAP, medications, allergies and diagnosis codes populated in the right fields on the right tabs.
