MedSim
An AI patient simulator that lets medical students practise difficult conversations in Norwegian, with a face, a voice and structured feedback.
- 24clinical scenarios
- 7 hourshackathon build
The problem
Most complaints against doctors are about communication, yet students get almost no practice. Trained actors are expensive and scarce, so a student rehearses two or three times a year and rarely on the hardest situations: breaking a serious diagnosis, uncovering hidden substance use, calming a frightened parent.
What it does
MedSim is a web app with a casebook of 24 scenarios across three difficulty levels. The student talks to the patient through the microphone. The patient listens, answers in natural Norwegian with a lip-synced video face, and hides facts the student has to uncover through good questioning.
When the consultation ends, a separate AI “senior consultant” reads the transcript and scores it on the five areas of the Calgary-Cambridge guide, checks which hidden facts came out, and writes concrete strengths and points to work on.
How it is built
The architecture builds none of the hard parts and connects three services with as little code as possible.
- ElevenLabs Agents is the patient’s brain, ears and voice. Each scenario is pure data injected into one shared prompt.
- HeyGen LiveAvatar or Simli provides only the face over WebRTC, swappable behind one interface.
- Claude writes the assessment afterwards. The patient model never grades itself, since it knows the hidden facts.
The server is deliberately thin: audio and video stream directly between browser and providers, and the backend only mints short-lived session tokens and runs the assessment.
Built in seven hours at the Mad Hack hackathon in September 2026, together with a pitch deck and a promo film.
