An outpatient consultation in a busy Indian hospital often lasts only a few minutes. In that time a doctor listens to the patient, examines them, decides what to do, and then has to write it all down: the history, the findings, the assessment, the plan and the prescription. A large part of a doctor's day goes into that record.
MedSum is a clinical documentation application built to take on much of that writing. It was built by Amrita Healthcare on BharatGen's open-weight AI models, and it is deployed and used by doctors and patients at Amrita hospitals. This case study looks at the problem it addresses, what it does during a consultation, and the AI models it runs on.
The problem doctors face
Three kinds of load come together in a single consultation.
- Documentation. Every visit needs a proper record. A doctor who types notes during the consultation splits attention between the patient and the screen. A doctor who writes them up later works from memory, often at the end of a long day.
- Language. Consultations in India rarely stay in one language. A patient may describe symptoms in Hindi or Malayalam with English words mixed in, while the medical record is usually kept in English. Someone has to bridge the two.
- Messy documents. Patients arrive with folders of paper: lab reports in different formats, old prescriptions, discharge summaries, photocopies and phone photos. Finding the one value that matters in that pile takes time the consultation does not have.
What MedSum does, step by step
MedSum supports the doctor before, during and after the consultation, and gives the patient something useful to take home.
1. Before the consultation: a summary from the patient's documents
The patient's medical documents are uploaded ahead of the visit. MedSum reads them using optical character recognition (OCR) built to cope with messy, real-world medical paperwork, and fills in structured fields from what it finds. Before the patient sits down, the doctor sees a short summary of the vital information.
The doctor can also ask questions of the documents directly. This is most useful for pathology reports, where the doctor often needs one specific result and should not have to page through every sheet to find it.
2. During the consultation: listening in real time
As the doctor and patient talk, MedSum transcribes the conversation in real time. Its speech-to-text has to handle medical terminology and the wide range of accents heard in an Indian hospital, not just everyday speech.
3. After the consultation: draft notes and a prescription
From the transcript, MedSum drafts a SOAP note, the standard clinical format that records the patient's Subjective account, the Objective findings, the Assessment and the Plan. It also drafts the prescription. When the consultation happens in Hindi, the transcript is translated into English and the same outputs are produced, so the record stays consistent whatever language the visit was in.
These are drafts. The clinical judgement, and the final say on the record, stay with the doctor.
4. For the patient: explainers in plain language
MedSum also produces patient-facing explainers of reports, prescriptions and discharge summaries. A discharge summary written for clinicians can be hard for a patient or their family to follow. An explainer sets out what it means in plainer terms.
The AI models underneath
Each step calls on a different kind of AI model. BharatGen builds all three kinds from scratch, for India, and makes them available as open-weight models that others can download, adapt and run.
| What MedSum does | Kind of AI model | BharatGen AI models |
|---|---|---|
| Transcribes the consultation as it happens | Speech-to-text | Speech |
| Drafts SOAP notes and prescriptions, translates Hindi transcripts into English, answers questions about documents, writes patient explainers | Language models | Text |
| Reads uploaded reports and prescriptions and fills structured fields | Document reading with vision language models | Vision |
Open weights matter in a hospital. Models whose weights can be downloaded can be adapted to a specialist vocabulary and run on infrastructure the user controls, an important consideration when the data is a patient's medical history. Amrita Healthcare's team built the application itself: the clinical workflow and the screens that doctors and patients use.
Languages
MedSum works in English, Hindi, Malayalam and Marathi. Transcription began with English and Hindi, and language support has widened since. For a hospital group whose patients speak many languages, the point is simple: a doctor should not have to choose between talking to a patient naturally and keeping a good record.
How the partnership works
BharatGen and Amrita Technologies signed a memorandum of understanding in December 2025 to work together on AI for healthcare, and MedSum was the first application taken forward under it.
The division of work is clear. BharatGen builds and releases the AI models. Amrita Healthcare, which knows clinical practice from the inside, built the application on top of them and deployed it in its hospitals. This is the pattern BharatGen wants to see more of: organisations with deep knowledge of a field building on open-weight AI models made in India.
More examples are on our ecosystem page. Our case study on Gyan Bharatam shows the same kinds of AI models at work on India's manuscripts.
