02 / Design Engineer and Founder

A safe space for medical students to learn from their mistakes.

Context

Teaching clinical reasoning with simulated clinics.

Medical school teaches the facts well and the judgement badly because judgement needs patients, and patients are the one thing a curriculum cannot hand out.

Pain

20students : 1patient

Fifteen minutes with one patient.
Live patients teach best, and logistics ration them.

Solution

AI plays the patient.

The student talks, examines, tests, treats — then sees where the reasoning broke, and goes again.

Outcomes

Unlimited patients

on demand
A case costs a doctor nothing to run, so a student can be wrong as often as it takes.

What I owned
  • Every function except the clinical one
  • The entire technology stack, written hands-on
  • Product — what got built, and in what order
  • All design and UX, from research to the interface
  • Sales and go-to-market
Team
Two founders: myself as CPTO, Dr Gopikrishnan Anjaneyan as CCO.
Stack

Svelte-TS · FastAPI · LangChain · RAG · LLM APIs · Figma

We wanted a way to make clinical reasoning visible and teachable — and give students a tool where they don't just receive knowledge, but build it through experience.
Dr Gopikrishnan AnjaneyanCo-founder · Associate Professor of Dermatology, AIMS Cochin

What we heard

Lack of real cases hurts medical learning.

Students describe theory they cannot apply and exams they do not feel ready for. Their teachers agree that cases are the answer, and cannot find the hours to build them.

  • I am not confident in my clinical exams and they are very stressful

  • Lectures are lengthy and boring

Student

  • We often have to refer 4000 page textbooks to make sense of the concepts

  • I don't know how this applies to the real world

  • Cases are the most important method of teaching I can think of but students don't get enough practice

Doctor

  • I don't have the time needed to prepare cases for my classroom teaching

Medical learning today

Only 20% is hands-on learning.

Lectures and textbooks build recall, and recall is what gets examined. The competence the job needs is built at the bedside — which is the smallest slice of the week, and mostly spent watching someone else decide.

Lectures and textbooks take four fifths of the week between them. The bedside — the only room where a student touches a patient — takes the rest.

What a case carries

What if every student and teacher had a patient?

Access to unlimited patient cases lets teachers teach and students learn through the lens of patient care.

  • History
  • Diagnosis

Patient case

  • Tests
  • Treatment

Teacher

Teaching using a patient case allows me to teach and explain concepts through the lens of patient care

Student

Applying what I learn helps me understand and retain the subject better

What we built

A virtual clinical environment with unlimited patients is better.

A virtual clinic lets students practise every stage of patient care — from history-taking to treatment — without real-world constraints.

Six stops on one loop. The closing edge is the product: nothing has to be scheduled or prepared, so the student simply starts again with someone new.

Inside the clinic

Case library, grouped by specialty
The consultation screen: a chat with the simulated patient, suggested openings beneath it, and a panel tracking which parts of the history are complete.
  1. The AI patient is prompted to answer in plain language and never give the diagnosis away.
  2. Students often did not know how to begin, so the screen suggests ways to open the conversation.
  3. A progress panel shows how much of the interview is done, and what is still missing.
Taking a history from the patient
A skin examination in the consultation: the findings written out above a clinical photograph of the patient's lower leg.A completed skin biopsy: the result written out above four stained slides labelled A to D.
  1. Requesting an exam returns a real photograph, not a written verdict.
  2. Results come back as images to interpret, so the judgement stays with the student.
Examination and lab tests, as in clinic
The submit-diagnosis dialog: each candidate condition carries a dropdown marking it primary, differential or ruled out, and the primary one asks for a written justification.
  1. An answer is not accepted until the reasoning behind it is typed alongside.
  2. Plausible wrong answers are mixed in, so every option has to be judged rather than skimmed.
Committing to a diagnosis and a differential
Three feedback cards: a starred evidence-gathering score, an accuracy card confirming the diagnosis, and a panel listing strengths beside areas for improvement.
  1. One score for how thoroughly they gathered evidence before deciding.
  2. Another for whether the call was right, with their reasoning quoted back to them.
  3. And the steps they skipped, so the feedback names the gap and not just the grade.
Feedback on how the case was reasoned
A numbered diagnostic timeline: ten steps, each tagged by task type — history taking, physical exam, lab test — with a one-line instruction beneath.
  1. The path is numbered, so the order an expert would have worked in is visible.
  2. Each step is tagged by type, so the kind of task reads before the words do.
  3. Heading first, detail under it, so the path can be scanned instead of read.
A suggested diagnostic timeline
A drug reference panel for clofazimine: why it suits this case, its indication and mechanism, dosing, a memory tip, alternatives, adverse effects and contraindications.
  1. A memory hook sits at the top, because recall is what gets tested.
  2. Why it matters here comes first, ahead of the reference detail.
  3. Warnings and cautions earned their place in user testing.
Drug concepts, read while treating
An OSCE question with four stain options, the chosen answer marked wrong and the correct one expanded with an explanation.The concept panel: an explanation of why AFB staining identifies leprosy, and three key concepts — specific, general and lateral.
  1. Opening the explanation teaches the idea behind the answer, not just which option was right.
  2. Questions are modelled on the real exam, and every option carries its own explanation.
Assessments shaped like the real exam
Walkthrough — a student works a case end to end
We thought students wanted to think like doctors — but they were trying to succeed as students. Success in exams was critical for that.

What moved the needle

Getting better at exams was the most important thing.

Students appreciated the learning, and valued retention and practice more. Shaping the assessments like the ones they actually sit raised engagement and outcomes together, rather than trading one for the other.

Aligning the learning design with what students were already measured on lifted engagement and effectiveness at once — the second line is steeper on both axes, not tilted between them.