AI at work
Doctors
Most doctors who use AI use it for the writing around the work: notes, referral letters, discharge summaries and the leaflet a patient takes home. That is where it saves time. It is not where the clinical judgement goes.
From Notes to Discharge: AI Workflow for Doctors
Prerak Juthani, MD
A doctor walks through the tools he uses across a patient’s stay, from the first note to the discharge.
Three tools to learn first
Everyday tools from Google, on the phone you already have. Learn these three and most of this page follows.
Gemini Notebook
formerly NotebookLM
Make a notebook of the guidelines you actually follow — the ICMR or society PDFs, your hospital’s protocols — and ask it questions. Every answer points to the page it came from. Then ask for a one-page patient leaflet or an infographic from the same sources.
Open Gemini NotebookGemini
with Canvas
Turn your notes into a presentation for the next CME or ward teaching in Canvas, or rewrite discharge advice in plain Urdu.
Open Gemini- Works offline
Gemma on your phone
in Google AI Edge Gallery
For anything with a patient’s details in it: it drafts on your phone, offline, and nothing is sent anywhere.
Open Gemma on your phone
New to these? Start with the short guide to all three, with a tutorial in Hindi and a note for Jio users.
Three things to try this week
- 1
A notebook of the guidelines you use
In Gemini Notebook, make one notebook and upload the guideline PDFs you actually follow. From then on the answers come from those pages, not from the AI’s memory.
In Gemini Notebook, upload the guideline PDFs, then ask: What do these guidelines say about [your question]? Quote the sentence, and say which document and section it comes from. If they do not cover it, say so. Then, in Studio: “Infographic” → a one-page patient handout on the same topic.
Check: Click the little numbers in the answer: each opens the exact passage. Read it. If the passage does not say what the answer says, the answer is wrong.
- 2
A referral letter from your own summary
Write the case as you would in a hurry — no name, no phone number, no ID. Let the AI turn it into a clear letter.
Here is a short case summary with no identifying details. Write a referral letter to a cardiologist: reason for referral, history, findings, current medicines, and the question I want answered. Keep it under 250 words. [paste your summary]
Check: Read every medicine and dose against your own notes. AI copies numbers wrongly more often than you would expect.
- 3
A leaflet a patient can actually read
Give it the instructions you already give out loud and ask for plain words at a Class 8 reading level.
Rewrite these discharge instructions for a patient who is not a doctor. Short sentences, plain words, a list of warning signs that mean "come back today". Do not add any advice I have not written. [paste your instructions]
Check: Look for anything it added. "Do not add any advice" helps, but it does not guarantee it.
In Kashmir
AI is already in some of the valley’s hospitals
- In 2023 the Directorate of Health Services, Kashmir approved AI chest X-ray reading (Qure.ai’s qXR) for district hospitals including Kupwara, Budgam, Shopian, Pulwama and Bandipora, and SDH Sopore — along with an AI-assisted ECG for heart attacks.
- If your hospital uses it, treat it as a triage flag that puts a film to the top of the pile — not as the report. The radiologist’s or your own reading is still the one that counts.
- Indian scribes are trained on Hindi and English. Before trusting one with a consultation that is half in Kashmiri, try it on a few of your own and read every line it writes.
Keep the trust
Take out the name, the number and the face. AI drafts; you sign.
A patient’s details are their health data. Once pasted into a chatbot, they are on someone else’s server, under someone else’s rules — and the note that comes back carries your name, not the machine’s.
- Before pasting: no name, phone, address, ID number, photo or rare detail that identifies the person.
- Use AI for documentation and explanation. Keep diagnosis and the prescription your own decision.
- If a scribe records the consultation, tell the patient and ask first.
The rules, in a line each
- Telemedicine Practice Guidelines (2020): AI may support a doctor but may not diagnose or prescribe on its own.
- ICMR Ethical Guidelines for AI in Biomedical Research and Healthcare (2023): a person stays accountable for every AI-assisted decision.
- DPDP Act (2023): a patient’s details are personal data — collect only what you need and keep it safe.
Tools
Examples to try, not recommendations. Where the data is sensitive, prefer the ones that keep it on your own computer.
For writing and summarising
Notes from the consultation (AI scribes)
Open source — the recording stays on your own computer
- OpenScribe Open source
Open-source scribe: records, transcribes on your machine and drafts a structured note. Needs some technical help to set up.
- Open Medical Scribe Open source
Open-source scribe that can run fully on your own machine, with no audio leaving it.
Reading the research
Clinic software, built in India
- HealthPlix
An EMR used by thousands of Indian doctors, with AI-assisted prescriptions in 11 languages.
- Eka Care
EMR for the clinic (EkaDoc), ABHA-linked records for patients, and the EkaScribe voice scribe.
Watch next
Indian Doctor Takes on Meta AI — Diagnosis Battle
Dr Avyact Agrawal, paediatrician
A professor of paediatrics puts an AI against his own diagnoses on real-style cases (2024, 40 minutes) — useful for seeing where it goes wrong.
How I use LLMs
Andrej Karpathy, a founding member of OpenAI
Not about medicine: one of the field’s leading researchers showing his own everyday use, tool by tool. It is two hours long — the first twenty minutes are enough to start.
Go further
- How doctors should use generative AI in clinical practice — safely and effectively
Offcall
From Dr Graham Walker, an emergency physician and the founder of MDCalc: AI for paperwork, not for steering decisions.
- How I Doctor (podcast)
Dr Graham Walker
Working doctors on what they actually use. Start with “AI secrets for clinicians”.
- NEJM AI Grand Rounds (podcast)
NEJM Group
Clinicians and researchers in depth — for when you want the evidence, not the tips.