About us

We're putting the conversation back into care.

India has built one of the most reachable and affordable healthcare systems in the world, and it runs on volume. Doctors see numerous patients a day, front desks handle phones, queues and billing at once, and hospitals serve entire districts. Care has scaled; the conversation around it hasn't. Patients speak many languages, often more than one in a sentence, and come to care unwell, anxious or unsure of what to ask. What slips through between visits, from instructions half-understood to medicines stopped early and reviews never booked, quietly undoes good treatment, and providers rarely get to see it.

IndRAH began as research into how patients describe what they're feeling before they see a doctor: in their own words, across languages, in real and imperfect conditions. That work showed us the hard part isn't generating language. It's listening, to hesitant and mixed speech, to clinical terms that speech recognition routinely mishears, and knowing when a person, not software, should take over. We still publish what we learn, including an open dataset on exactly that problem.

What grew out of that research is our conversational framework for care providers, which answer patients' calls and follow up after visits, and the software that runs the day around them for front desks, doctors and labs. The agents work within firm limits: they say only what a doctor has signed, never diagnose, and hand over to a person whenever it matters. We build from inside clinics and hospitals, with clinicians and researchers alongside us, starting with patient management and journey mapping while growing into nuanced specialty and chronic care management.

Our name says what we set out to build: Indigenous Real-time Agents for Healthcare, made in India for the way India speaks and seeks care. The aim behind it is simple. No patient should drift out of their own care because there was not enough time to talk to them.

Research

We publish what we learn.

The people who reach out for care are often unwell, tired or worried, and they speak the way India speaks: in many languages at once, with medicine names no speech model was trained on. We study that openly, and share the methods and data.

Latest release · Dataset

IndMedSTT-Synth

When speech-to-text mishears a medicine, the record breaks. Our first public dataset maps exactly how, for Indian multilingual speech.

Open on Hugging Face →

Build it with us.

Whether you run care, invest in it, or want to work on it, we'd like to hear from you.