FR
In use Health education

CareConverse – Conversation practice with AI personas

Health students practise difficult conversations by voice with an AI-played patient, relative or colleague. Teachers write the scenarios without code, and the platform follows the learning objectives as the conversation unfolds.

Why

Communication is at the heart of care: breaking bad news, taking a history, handing over a patient, calming a worried relative. These skills are hard to practise, because simulated patients and supervised role plays are costly and rare. CareConverse lets students rehearse such situations as often as they need, and gives teachers a way to design them and see where their class stands.

A student's voice session: the patient's portrait and name, tabs for the other people in the scene, the conversation transcript, and on the left the objectives reached so far and the medical record.
A voice session with a simulated resident of a care home. Left: the objectives reached so far and the documents for the case. Demo scenario; the interface is in French.

What it does

  • Real-time voice dialogue

    Students speak naturally with the simulated person, or type if they prefer.

  • Group conversations

    Several characters, each with their own voice, who also talk to each other, for example during a coffee break.

  • Live objective tracking

    The communication objectives a student demonstrates are ticked off during the conversation.

  • Guided steps and hints

    Scenarios can unfold in steps, with hints that become more explicit when a student is stuck.

  • Feedback and progress

    Feedback at the end of each session, and a dashboard of past sessions and progress.

  • Authoring without code

    A step-by-step editor for the situation, the personas, objectives, steps, triggers, references and languages.

  • Test mode

    Teachers can play their own scenarios without anything being recorded.

  • Classes and analytics

    Teachers assign scenarios to classes and follow results across a cohort.

  • Several languages

    The interface runs in French, German and English, and scenarios can be translated.

The teacher's scenario editor on its Situation step, with the instructions shown to the student in a green panel and the hidden instructions for the AI in a red panel.
The scenario editor. Teachers decide what the student sees (green) and what only guides the AI character (red), such as how the person reacts when asked for a pain score.

How it works

The teaching content is not written in the code: every scenario, persona and objective is data that teachers edit. During a conversation, each student turn is handled twice in parallel: one model answers in character, while another checks which objectives the student has just demonstrated. In group scenes, a "scene director" decides who speaks next. The microphone is only used after the student has agreed to it.

A student's dashboard with the number of sessions, scenarios assigned, objectives reached and practice time, the scenarios available, a progress chart per attempt and the list of past sessions.
The student dashboard: practice time, objectives reached, and progress from one attempt to the next.

Technology

  • Web application: Vue 3, Node.js and Express, PostgreSQL, deployed with Docker on HES-SO infrastructure.
  • Language and voice: Google Gemini models, including Gemini Live for full-duplex voice conversation (cloud services).
  • Speech recognition can also run locally with faster-whisper.
More

Other projects

IPC4MH DiverSSity XR4Inclusion Industrial guidance Greta XRXP