Every student has the difficult conversation, before it counts.
A student talks to an AI character you define: a supplier who will not move on price, an employee who disputes a review, an investor with ten minutes. Afterwards the student gets a score on your criteria, and you get the transcript behind it.
Text, voice or video avatar · English and Spanish · in the browser
Decide with numbers. Speak with a person.
No AI while the student plays
Prices, capacity, budgets, rounds. A deterministic model works out the result: the same decisions always give the same outcome, and no student data goes to an AI provider.
See the catalogueThe AI is the other person
Some skills only show when someone answers back. Here the model plays a character in a conversation, and a second pass scores the whole dialogue. It uses AI while the student speaks, and this page says exactly where.
From a situation to an evaluated conversation
We set up the scenario with you
You describe the situation, who the student talks to, how difficult that person is and what matters most. During early access we build it, and you approve it before students see it.
The student has the conversation
In the browser, alone, for up to about nine minutes in real-time voice. The character listens, pushes back and follows up on what the student actually said.
Each student gets an evaluation
Five criteria with the weights you set: understanding, communication, empathy, professionalism and solution focus. With strengths, what to improve and the full transcript, so you can check a result before it counts.
You choose the mode for each scenario
The same scenario and the same evaluation behind each one.
Text chat
Typed, turn by turn. Works on any device and any connection.
Turn-by-turn voice
The student speaks and the character answers aloud, one turn at a time.
Real-time voice
A spoken conversation with no waiting between turns. The student can interrupt, as in a real one.
Video avatar
Real-time voice with a face on screen; here the student speaks once the character has finished. Available on request, because it adds a video provider.
Six kinds of conversation
These are the scenario types the platform supports today. The examples show the kind of situation; yours is written for your course.
Designed a role-play in Primer?
Primer, the open-source project from Wharton Generative AI Labs, helps an educator design a role-play through an interview. To use it with a class, each school deploys its own copy and funds its own AI account, and its authors describe it as an experimental prototype.
We import your experience: each character becomes a scenario with your instructions, running here with your class list and an evaluation per student. Send us the folder as a .zip and we review the result with you before your students see it.
Send us your Primer experiencePrimer is a project of Wharton Generative AI Labs. Eureka Simulations is not affiliated with it.
Where this stands today
We would rather you knew now.
- Early access. There is no self-service sign-up. We set up your first scenarios with you.
- English and Spanish. The student interface is in these two languages today.
- AI while the student speaks. Depending on the mode, the conversation is processed by Anthropic, OpenAI, ElevenLabs or HeyGen, which receives the character's voice only; in turn-by-turn voice, the student's browser does the speech recognition. We keep the transcript and record no audio or video. Ask us for the sub-processors of the mode you choose.
- No automatic gradebook sync. Evaluations stay on the platform for you to review; today they are not sent to your LMS gradebook.
- One character per conversation. A scene with several characters taking turns is not available yet.
- Priced by quote. Each conversation has an AI cost, which we measure per attempt; the quote depends on the mode and the number of students.
Frequently asked questions
Is this for practice or for grading?
Start with practice. The evaluation applies the same criteria and weights to every student and shows the transcript behind each score, so an instructor can check it. For a grade that counts, we recommend that a person reviews the result first. The same engine runs oral exams in our assessment product, Pro Evaluation System.
Do I have to write prompts?
No. You tell us the situation, the character, the difficulty and what matters. During early access we build the scenario and you approve it before students see it.
What happens to my students' voices?
Nothing is recorded: no audio and no video. The platform keeps the transcript and the evaluation. In real-time voice, OpenAI processes the speech to hold the conversation; in avatar mode, HeyGen receives the character's voice only.
How long is a conversation?
You set it per scenario. Real-time voice conversations run up to about nine minutes: long enough for one difficult exchange and short enough to repeat.
I built an experience in Primer. Can I run it here?
Yes. Send us your experience folder as a .zip (experience.json and resources/) and we import it: each character you pick becomes a scenario with its instructions, and the objectives of its phase become what the evaluation rewards. Before your students see it, we check each character's opening line with you. Scene images, voices and the report agents are not carried over, and each character becomes a conversation of its own: Primer's hand-offs between mentor, practice and feedback are not kept.
How does this fit with your decision simulations?
They are two separate tools on the same platform. A decision simulation uses no AI while the student plays; role-play does. Some courses use both: students decide in the simulation, then practise the conversation that decision leads to.
Try one conversation with one class.
Tell us the course and the situation. We set up the scenario, your students have the conversation, and you read the evaluations.
Request early accessLast reviewed: 2026-11