Reading an interview answer aloud does not reproduce an interviewer questioning it. Rehearsing a presentation does not reproduce someone interrupting to challenge an assumption. You can practise your words alone, but recreating the people and the flow of the conversation is harder.
Vime was an AI online-meetup simulator for practising with configurable participants and a configurable session. I was solely responsible for product planning and design, and took the project through a prototype. The intended uses included interviews, language practice, customer conversations and investor pitches. My work centred on how a user would construct a situation, navigate the conversation and review it afterwards.
Designing the participants and the sequence
An interviewer and an investor have different reasons to listen to an answer. A supportive participant and a demanding one can bring different pressures to the same conversation. Putting all those conditions inside a single prompt would make them difficult for the user to inspect and change.
The creation interface separated them into Agenda, Personas and Layout. Agenda described the topic and purpose. Personas defined participants’ roles and personalities. Layout described the session’s sequence and timing. Here, “layout” meant the structure of the conversation: who would speak during each segment, how long it would last, and whether responses would proceed automatically or under the user’s control.
For an investor-pitch rehearsal, a user could separate the presentation from the question period. They could keep the pitch while changing a participant’s role or temperament. The aim was to make the conditions editable without asking the user to describe the whole exercise again.
Giving the conversation a meeting interface
I designed the conversation screen around the form of a video meeting. Participants were visible together, the current speaker was identifiable, and a sidebar showed the agenda and active segment. Conversation history, microphone and keyboard input, pause and end controls belonged to the same flow.
A presentation also involves material that is not spoken. File upload and screen sharing were planned so the participants could discuss a shared slide or document. Researching a real person to help construct a persona was another part of the proposed preparation flow. These remained product and design scope, rather than verified production features.
The prototype included simulation creation and storage, persona configuration, conversation records and a speech-input path. Its data model separated the simulation, reusable personas, and ordered segments with durations. The conversation code maintained a current persona and message history, with a flow for moving to another persona when the current exchange ended. It made the basic mechanics of a conversation with several participants concrete.
What would remain after the meeting
Report was designed as a place to revisit the session. The proposed view combined speaking rate and pronunciation indicators, relevance to the topic, and feedback from each persona with the conversation record. The design aimed to help a user find a specific statement to work on, rather than leave them with a score alone.
The exercise itself could also be reused. I designed Explorer for discovering simulations and adding them to a personal library. Public and private visibility, authorship, library ownership and reusable personas in the code reflected that direction: a useful rehearsal configuration could become something another person started from.
What the prototype established
The work produced a product structure, interface designs and a prototype for exploring that structure. The purpose of a conversation, the people in it and its sequence became separate things a user could edit. That was the common model for the different rehearsal scenarios.
The prototype’s report used example metrics and feedback; it did not establish learning gains or pronunciation-assessment accuracy. The six-language launch in the deck was a roadmap item. The contribution was turning the conditions of conversation practice into an interface users could configure, with a prototype that made the intended interactions tangible.