For clinicians

We are asking you to evaluate this, not to use it.

There is no clinically validated product here. There is a working prototype, a published list of its defects, and a research programme that needs clinician judgement to mean anything. This page explains exactly what participation involves.

Who this is for

Kerala clinicians and clinical researchers.

01

Practising clinicians

Physicians and surgeons who document consultations daily and can say, immediately, when a note misrepresents what was said.

02

Clinical researchers

People who have built or critiqued evaluation protocols and can tell us where ours is weak before we run it.

03

Medical educators

Those who teach documentation and can judge whether a structured draft teaches or misleads.

What participation involves

Three levels, increasing commitment.

  1. A demonstration — 20 minutes

    A live session on synthetic cases. You choose the inputs. Text and audio paths are shown separately and labelled. The known defects are reproduced on request.

    Commitment One conversation. No data required.

  2. Structured evaluation — a few hours over weeks

    You receive a written protocol, the synthetic case set, and a severity rubric. You score outputs independently and return your judgements. Disagreements between evaluators are recorded, not averaged away.

    Commitment Your clinical judgement on record, for a named evaluation.

  3. Reference corpus contribution — a research collaboration

    The most valuable input, and the most carefully governed: consented recordings or transcripts from your own practice, with your own gold-standard documentation, under a written protocol and an ethics process you approve.

    Not yet open The protocol and consent framework are not finalised. We will not accept clinical recordings until they are.

The one thing we most need

Gold-standard documentation.

Right now "correct" means an engineer compared the output to what they expected. That is not evaluation. It is the reason no accuracy figure appears on this site.

What would change it:

  • Consultation transcripts or recordings you have consented and de-identified.
  • Your own documentation of what the correct note should contain for each.
  • Permission to test against them, and to publish aggregate results including failures.

With that, a claim becomes possible. Without it, every statement about quality remains an engineering opinion, and we will keep describing it as one.

What you keep

  • Your transcripts remain yours.
  • Nothing is published without separate written agreement.
  • You can withdraw and have contributed data deleted.
  • Your name is not used in any public material without consent.

What we will not ask for

Patient data through this website. Anything unconsented. Anything you cannot lawfully share. If a demonstration is the right starting point, no data is involved at all.

Honest expectations

What you should not expect from us.

A validated product
There is none. Eight open defects are published on the research page.
Malayalam capability
Malayalam is research. Code-switched input currently fails, and one case inverts meaning. See the Malayalam page.
Free software
This is not a free tier or a beta programme with an upgrade path. It is a research collaboration with no commercial commitment on either side.
Institutional endorsement
No hospital, university or regulator has endorsed this work. Any future affiliation will be stated explicitly, with permission.

Start here

Twenty minutes, synthetic data, no commitment.

If it is not worth your time after that, we would rather know.

Request a research demonstration Read the defect register first

Please do not send patient data through this website or by email. Enquiries only.