How HealthPlix turns doctor consultations into medical records with Sarvam
"Clinical documentation only works if it understands how doctors actually speak. In India, that means multiple languages, regional accents, code-switching, clinical shorthand, and the reality of a busy consultation room.
With Sarvam, we’ve embedded multilingual transcription into HALO that can capture these conversations more naturally. It helps doctors spend less time on documentation and more time with patients.”
Chaitanya Raju
Executive Director and Chief Product Officer, HealthPlix
Background
HealthPlix is India's largest AI-powered Electronic Medical Record (EMR), used by more than 14,000 doctors across 1.5 lakh outpatient consultations every day. As Healthplix scaled, making the EMR genuinely usable inside a real consultation became central to its product direction. HealthPlix began embedding AI across the consultation workflow, with a focus on voice-driven documentation that could capture conversations naturally, with doctors as the primary users and medical scribes positioned as a potential extension of the same workflow over time.
When documentation gets in the way of care
A doctor-patient interaction in India carries more complexity than most documentation tools are built for. Doctors switch between English, Hindi, and regional languages mid-sentence, refer to medicines by Indian brand names, and use specialty-specific shorthand. In a busy clinic with background noise and multiple people in the room, capturing all of this accurately is not straightforward.
Most ASR systems are not built for this. Accents cause errors, Indian drug brand names come out mangled, and code-switching falls outside what standard models are trained on. When transcripts are unreliable, doctors correct and rewrite, which defeats the purpose entirely. HealthPlix built HALO to fix this. Getting it right meant finding AI that truly understood how Indian doctors speak.
Why Sarvam
HealthPlix evaluated multiple frontier AI models against real consultation audio before settling on a partner. Sarvam consistently outperformed the alternatives on the dimensions that mattered most in an Indian OPD: prescription accuracy, interpretation of regional dialects, and the ability to follow code-switching and clinical terminology in context.
Sarvam's biggest differentiator was accuracy in the Indian context. While HealthPlix's proprietary models handled the clinical layer of mapping transcripts to structured records, drug names, and terminology, Sarvam's Speech to Text model brought deep understanding of regional languages and accents, which was crucial to achieving the required accuracy. The Integration was straightforward, with well-structured APIs that reduced the customisation required to get there. Sarvam's responsiveness and flexibility in supporting Healthplix’s healthcare-specific workflows helped maintain the standards a clinical environment demands, while its alignment with HealthPlix's compliance and security standards supports the company's commitment to protecting patient data.
Inside a consultation with HALO
HealthPlix deployed Sarvam's Speech to Text model, Streaming V2.5, within HALO. Sarvam transcribes the consultation in real time, and HealthPlix's models, trained on years of clinical data and domain expertise, structure that transcript into a complete clinical record covering vitals, complaints, diagnosis, medication, advice, and treatment plan. The same workflow extends naturally to medical scribes, where clinics choose to use them.
To illustrate how this works, consider a typical consultation. Doctors often set aside clinical terminology to help patients understand what they are being told. A doctor in a diabetes practice might say "your sugar is on the higher side" to put the patient at ease, while intending to record "Type 2 Diabetes Mellitus" in the prescription. With Sarvam's Speech to Text model capturing exactly what the doctor said and HealthPlix's extensive model training interpreting the clinical meaning behind it, the structured record reflects the terminology accurately. With documentation handled automatically, every consultation ends with an accurate clinical record that the doctor can review.
The Impact
- Accuracy: 97%+ accuracy on prescriptions generated through HALO, enabling doctors to review rather than rewrite.
- Time saved: Approximately 5 minutes saved per consultation, recovering the time doctors previously spent typing prescriptions.
- Adoption: More than 50,000 consultations completed on HALO to date.
- Patient experience: In a country with a low doctor to patient ratio, every minute saved on documentation goes back to the patient, meaning more time listening, more time explaining, and better care for every patient who walks through the door
- Latency and cost: Improved transcription speed and reduced costs relative to alternatives, giving HALO headroom to scale without compromising accuracy.
Looking Ahead
Over the next 12 months, HealthPlix expects its use of Sarvam within HALO to deepen across expanded language and dialect coverage, higher per-doctor consultation volume, and broader specialty adoption, with longer-term exploration of other voice-led use cases. The partnership reflects a shared commitment to building AI that is genuinely fit for the Indian clinical context, and as both organisations continue to advance their capabilities, it is set to play an increasingly meaningful role in the future of AI-led healthcare in India.