Project case study
Vivalingo
I wanted language practice to feel closer to a conversation I might actually have. Vivalingo combines useful phrases, recall practice and an AI speaking tutor with a review schedule that keeps bringing important material back.
My role: Product direction, learning-flow design and iteration with AI-assisted development.
Watch the interactive walkthroughHow it works
Practice a useful phrase
A lesson introduces language for a concrete situation.
Typed course content supplies phrase and exercise data. The lesson interface presents practice without requiring a model call for every card.
Recall before revealing
The learner attempts an answer before checking it.
Exercise state tracks the attempt and response. Revealing the answer is separate from generating a new AI explanation.
Schedule the next review
Difficulty ratings change when the phrase comes back.
The review scheduler maps again, hard, good and easy ratings plus review count to day intervals, while tracking strength and lapses.
Speak with the tutor
A staged conversation shows the voice flow.
Browser WebRTC carries audio to the AI service; a server authenticates and supervises the session. Caption-aware turn handling coordinates responses after interruptions.
Practice the part you will actually need.
The product moves between structured practice and open conversation. Lessons give the learner something concrete to work on; review brings phrases back later; the tutor creates conversational opportunities to use them. I have iterated the voice experience around interruptions, captions and keeping track of whose turn it is.
- Focus
- Typed lesson content, review scheduling, WebRTC and server-mediated AI sessions
- Status
- Personal learning product; continuing voice and mobile iteration
How it works
Typed course content supports structured practice. A deterministic scheduler maps review ratings and review count to intervals. For voice, WebRTC carries audio while a server authenticates and supervises the AI session and coordinates caption-aware turn handling.
A decision that mattered
The audio and the caption are related but not identical. Speech recognition can miss a phrase even when the voice model heard it. The voice workflow waits briefly for captions, keeps audio authoritative and handles an interrupted response before starting another one.
What comes next
I am continuing to refine the voice flow and mobile experience. The walkthrough uses fictional learner data and does not claim measured fluency gains.