- Healthcare
- Updated
Clinical Notes by Voice: 35+ Hours a Month Back per Clinician
Administrative work on notes and referral letters was costing doctors hours every month. We built an AI assistant on Azure AI Foundry that transcribes consultations in 19 languages and generates referral letters, returning more than 35 hours a month per clinician, with documentation errors down 60% in post-visit audits and 80% clinician adoption within four weeks.
Book a 30-Minute Working Session
The situation
Physicians spent too long typing notes and letters.
In busy clinics, physicians spent too long typing notes and referral letters, and patient education and follow-up slipped. Low-quality or delayed documentation risked clinical errors and compliance gaps, and tools without conversational or multilingual support went unused in diverse practices. The need was to automate documentation without compromising care quality or the strict privacy standards that clinical data demands.
What we built
A voice assistant that turns consultations into structured notes.
We built an AI-first assistant on Azure AI Foundry. Foundry speech models transcribe consultations in real time across 19 languages, and the transcripts are structured into notes. GPT-4 then suggests diagnoses and next steps for the clinician and generates referral and patient-education letters from therapeutic guidelines, while a Power BI Copilot interface answers clinicians' natural-language queries. Responsible AI is designed in: voice is ephemeral and encrypted, no patient health information is stored after a session, access runs through Azure AD and Purview, every output carries a confidence score and source provenance, and the solution scales on AKS.
How it works
The architecture, stage by stage.
Capture
- Voice capture · consultations in 19 languages
- Azure AI Foundry Speech · real-time transcription · structured notes
Assist
- GPT-4 · diagnoses and next steps · referral and patient-education letters from therapeutic guidelines
- Power BI Copilot · natural-language clinician queries
What changed
The outcomes, as the client measures them.
- More than 35 hours a month returned per clinician.
- Documentation errors fell 60 percent in post-visit audits.
- Referral letters are generated 75 percent faster.
- Note-transcription accuracy is above 90 percent.
- Clinician adoption reached 80 percent within four weeks.
- Doctor satisfaction rose from 6.5 to 8.7 out of 10.
- No patient health information is stored after a session.
By the numbers.
35+ hours
Returned per clinician each month
60%
Fewer documentation errors in post-visit audits
75%
Faster referral-letter generation
80%
Clinician adoption within four weeks
Facing the same problem?
Start with a working session.
Bring your version of it to a 30-minute working session. We will walk through how this was built and what it would take on your data.