Your history,
ready before you
sit down.say hello.take a seat.walk in.
आपकी बात, आपकी भाषा में — डॉक्टर तक पहले पहुँचे।
MediKiosk listens to the patient in their own language, reads the papers they already carry, and puts a verified clinical history on the doctor's screen before the consultation begins.
A history the doctor can trust. Not a form the patient guessed at.
Registration captures a name and an age. Health apps assume a smartphone and a login. Neither reads the crumpled prescription in the patient's hand. MediKiosk does both, in the patient's language, and reads it back before the doctor sees a word.
Speak or tap, in seven languages
Hindi, English, Marathi, Gujarati, Punjabi, Tamil, Telugu — the whole interview, every option read aloud. Follow-ups written for the complaint, the way a physician would ask.
The paper they already carry
Prescriptions, lab reports, discharge summaries — photographed at the kiosk and read line by line: diagnoses, medicines with doses, values against their ranges, in date order.
An AI-summarised consultation, verified by the doctor
Answers and scanned papers become one structured clinical summary — chief complaint to review of systems, ICD-10 coded — read back to the patient in their language, then accepted or amended line by line by the doctor, with interaction warnings as they prescribe.
The history is taken before the door opens.
What the flow costs and what it gives back, from the demo cases.
Prototype estimates from the demo flow with the example cases. Not yet measured in a live OPD.
Six steps. The patient does the first five before they sit down.
Each one is a screen at the kiosk, in the patient's language, read aloud if they want.
Choose a language
Seven scripts on the first screen. From here on every word, prompt and option is in that language.
Patient · kioskSay who you are
Mobile, ABHA number or Aadhaar — a one-time code proves the phone. Returning patients are recognised; new ones get an ABHA and, if they like, a login.
Patient · kioskAnswer by voice or tap
Chief complaint, then follow-ups written for it. Point to where it hurts on a body map. An Ayurvedic history if the hospital asks for it.
Patient · kioskUpload your documents, get your token
Hold each old report or prescription to the camera — it is read, dated and filed into a timeline, abnormal values flagged. Then a token slip prints with a QR code and your department.
Patient · kioskYour AI summary, read back to you
Everything you said and every paper you scanned becomes one clinical summary. The kiosk reads it back in your language; you confirm it and choose what the doctor may see — before anyone else reads a word.
Patient · kioskThe doctor verifies and prescribes
The structured history is already on the console, sources beside every line. Accept or amend, see interaction warnings, push a FHIR record to the ABHA locker.
Doctor · consoleEvery module the problem statement asks for.
Built means working in this deployment. Partial means working with a stated limit. The full list, including what is mocked and why, is on the about page.
Adaptive questioning
Four to five follow-ups written for the chief complaint — SOCRATES elements, tap-able options, "asked because you mentioned…".
BuiltVoice, both ways
A mic on every open question and spoken next / back / repeat. Every screen reads itself aloud, in all seven languages, on any machine.
BuiltAyurveda history mode
Dashavidha Pariksha, Agni, Koshtha, Ahara-Vihara — all ten parameters, Vikriti kept apart from Prakriti.
BuiltRed-flag triage
An emergency button or a red-flag answer: the token turns priority, the case tops the queue, the console sounds, staff get an SMS and email.
BuiltReads the paper
Printed and handwritten prescriptions, lab reports, discharge summaries — read, extracted, dated into a timeline, abnormal values in red.
BuiltDrug interaction check
Scanned papers, reported medicines and today's prescription as it is typed — curated rules for allopathic and Ayurvedic pairs.
BuiltConsent first, revocable
Four separate permissions, read aloud. "Withdraw consent" on the patient's dashboard locks every clinician out of that visit.
BuiltFHIR R4 record
A real document bundle — Composition, Condition with ICD-10 and NAMASTE codes, Observations, Consent — downloadable from the console. The push awaits ABDM sandbox access.
PartialThe patient keeps a copy
A login ID sent by SMS and email; "My records" shows every visit, document, diagnosis and prescription, read-only, for good.
BuiltWhat changes when the history is taken at the kiosk.
Same patient, same doctor, same ten minutes. What each side gets to work with.
A kiosk that speaks your language.
Big buttons, a voice for every screen, and nothing to read if you would rather listen. About four minutes.
Open the kiosk For the clinicianA console that has already done the asking.
The queue, the structured history with its sources, the scanned papers in date order, and a prescription pad that warns as you type.
Open the consoleThe things you may ask first.
Is MediKiosk a medical device? Does it diagnose?
No. It takes a history and organises documents. Every clinical decision is the physician's; the console shows a draft to accept, amend or reject, and records who did which. It is a Smart India Hackathon prototype and not a live service.
Where does the patient's data go?
Voice never leaves the kiosk as audio — recognition runs in the browser and only text is kept. Document and summary text go to a model API over TLS. In a hospital deployment the same code points at a hospital-hosted model, so nothing leaves the building. Consent is granular and revocable from the patient's own dashboard.
Which languages, and how deep does it go?
Hindi, English, Marathi, Gujarati, Punjabi, Tamil and Telugu — the whole interview: every question, option, consent line and error message, plus spoken prompts. The physician's summary is always in English; the read-back to the patient is in their language.
Does it work when the Wi-Fi drops?
The kiosk shell — pages, scripts, language packs — is cached, so the screen never goes blank. Anything that needs the server (codes, summaries, the queue) waits for the network and says so, rather than showing stale data.
Are the ABHA numbers real?
Not yet. They are validated by format and generated locally; issuing real ones needs ABDM sandbox credentials, which are applied for. The FHIR R4 bundle that would be pushed to the ABHA locker is real and downloadable today — only the transport is mocked.
What if the patient cannot read, or cannot see well?
Audio-guided mode reads every screen, heading, hint and option aloud, and accepts spoken answers and spoken navigation. Large-text and high-contrast switches are on the accessibility screen. A screen left alone clears itself within two minutes.
Ready to see a visit run end to end?
Three minutes as a patient, then open the console as the doctor. The example cases load from the console if you want a full queue.