Practice · Step 7
Live interviewer: Agent 01 and Agent 02
The Live interviewer is a spoken practice interview. Agent 01 is a JB-Live interviewer; it asks, you — or Agent 02 — answer. With Agent 02 you can watch a full interview answered from your own notes before you give it yourself.
Who answers Agent 01
| Choice | What happens |
|---|---|
| I answer (microphone) | Agent 01 asks, you answer into the microphone and Agent 01 hears you |
| Agent 02 (separate JB-Live) | A second JB-Live with its own voice hears Agent 01 directly; its answers fill the Answer window |
| Through the app | The interview's own JB-Live answers Agent 01 from system audio, like a real call |
Run a practice interview with Agent 02
- Open the Live interviewerTraining → Practice → Live interviewer → Open.
- Choose Agent 02Select Agent 02 (separate JB-Live).
- Write the opening lineAgent 01 says it word for word. Leave it empty and Agent 01 greets however it likes.
- Check the routingThe line under it shows where each agent speaks and the role, for example Agent 01 → System audio · Agent 02 → Microphone · interviewing for Senior software engineer. Change it in Devices → JB-Live agents.
- Start trainingAgent 01 opens. Agent 02 is briefed from your notes before the first question and searches them while each question is asked.
- Stop trainingStop when you have seen enough; the sessions close at once.
A real run, one minute
A recorded 60-second run of the installed app: Agent 01 asks its whole question, then Agent 02 answers from the notes without being cut off. The Answer window on the right fills as Agent 02 speaks, each part marked grounded.
Taking turns
Two AI voices would talk over each other if left alone. JBWhisper keeps a floor: whoever is speaking keeps it until their last spoken word has played out and 700 ms of quiet have passed after a finished sentence (a pause mid-sentence is a breath, not the end of a turn); Agent 01 always opens; what the other agent says meanwhile is held and played right after, in order. Only speech counts — the silence a voice model streams between turns does not keep anybody on the floor.
Both agents are live voice sessions that are billed while they run. A minute or two shows a full exchange; stop the training when you have what you need.