Hyperlink Dynamics

Interactive lesson demo

Scripted prototype

Meet Miss Lumi, an AI speaking tutor, in a 3-minute voice conversation lesson for ages 5–9: “What’s the weather like?” She talks, the microphone opens on its own, and the child answers out loud. The AI teacher and the courseware (the lesson screen) work together through a two-way lesson protocol, and every message between them shows up in the protocol log.

🎧 Best experienced in Chrome or Edge, with your microphone and speakers on. This page runs a scripted lesson policy in your browser. In the product, an AI reasoning layer (a large language model such as Claude or GPT) generates the teacher’s moves in real time. How it works

Watch mode plays a simulated student who speaks the answers, so you can see the whole loop without a microphone.

Courseware

    What’s the weather like?

    A 3-minute voice conversation lesson · ages 5–9

    🎧 Turn your speakers on. Best in Chrome or Edge with a microphone.

    AI teacher

    Miss Lumi

    AI speaking tutor · scripted policy in this demo

    Ready when you are

    Hi! Press Start lesson and talk with me, or turn on Watch mode to see a simulated student take the lesson.

    Protocol log

    0 downlink commands, 0 uplink events

    ⬇ teacher → courseware ⬆ student → teacher

      Messages appear here as the lesson runs: commands going down to the courseware, events coming back up from the student.

      How this demo works

      • A voice conversation, run on a screen. The AI teacher speaks and decides what happens next. The courseware shows it, opens the microphone when it’s the child’s turn to speak, and reports what the child says or taps. The two only talk through small JSON messages, and every one appears in the protocol log.
      • Commands go down, events come up. Downlink commands such as show_cards, play_model, open_mic and award_star drive the screen. Uplink events such as speech_result, card_tapped and request_repeat tell the teacher what happened.
      • The teacher adapts turn by turn. A near miss gets a syllable hint and one more try. A second miss gets a kind model and the lesson moves on. A one-word answer gets scaffolded into a full sentence. Nobody is ever told “wrong”.

      Microphone: works best in Chrome/Edge. Speech is recognized by your browser’s built-in speech service; this demo does not record or store audio on our servers. (Some browsers send audio to their own speech service to recognize it.) The teacher’s voice also comes from your browser, so turn your speakers on. If you can’t use a microphone, a small “Can’t use a microphone?” option under the mic lets you answer with buttons instead. Privacy policy

      Protocol reference (generated from protocol.js, the same schema the demo enforces)