# Live session

> Connect audio, run a session, answers as you go.

## What it does

Cariara listens to your interviewer’s audio, transcribes their questions in real time, and streams a structured answer from Sona — your AI interview helper — into a side panel only you can see. You read it, internalize the structure, then answer in your own voice. Four answer surfaces, each tuned for a different interview type: Home — the live session surface. Sona answers questions in real time as the interviewer asks them. Coding — multi-language playground with three approaches (brute / optimized / most optimal), narration, complexity analysis, auto-run, and a follow-up Sona sidebar. Design — capacity estimates with an interactive Scale Calculator, requirements, scalability tiers (clickable drill-downs), architecture diagram, trade-offs, and edge cases. Behavioral — STAR-format response personalized from your resume and job context.

## Before the call

Open /live in a browser tab separate from your video call. Run the Audio Setup Wizard — it appears automatically the first time. It detects how to hear the interviewer (desktop app, virtual cable or meeting tab); you confirm the mic and connect. Run Audio Check from the icon-rail menu. The modal verifies mic permission and lets you confirm the capture path with a live level meter. It’s a quicker test than the full wizard — do it 15 minutes before the call. If you’re using room-mic or mic-only capture, finish Voice enrollment first — it takes 15 seconds and prevents Cariara from transcribing your own voice into Sona.

## Audio setup

The Audio Setup Wizard detects how to hear the interviewer and saves the result to your account. What it picks: Desktop loopback — Cariara desktop app captures system audio natively (any meeting client, no virtual cables). Tab share — share a Chrome tab with Share tab audio checked. Works for Zoom Web, Meet, Teams Web. Chrome / Edge only. Virtual loopback — BlackHole (macOS) or VoiceMeeter (Windows) for capturing native Zoom / Teams desktop clients. Mic-only fallback — lossy, speaker diarization filters your voice from the stream. Use only when nothing else works. Full walkthrough for each method, with troubleshooting, lives at Audio setup .

## The four tabs

Switch tabs before the question lands so the right surface is loaded. The Home tab is where the live-interview Sona stream lives; Coding / Design / Behavioral are standalone surfaces with their own input boxes you can drive directly (typed, voice, image drop, or fetch-from-URL). When to use it What you get Default — Sona answers the interviewer’s questions live as your audio capture streams. Streaming answers in a side panel, screenshot capture, session history. A specific algorithm / data-structure problem. Useful before the call to drill, or to manually drive Sona. Three approaches with complexity, auto-run, voice-input, drag-drop image, fetch-from-URL, follow-up Sona sidebar. A "design X" or "scale Y" question. Drive it manually or let Sona auto-route from Home. Requirements, Scale Calculator, scalability tiers (clickable drill-downs), architecture diagram, trade-offs, edge cases. STAR-style or motivational questions. Situation / Task / Action / Result, personalized from your resume and target company.

## Using Sona during the interview

Sona’s answer streams in token-by-token. Start reading as it appears — you don’t need to wait for the full response. Sona answers in first person as the candidate ( I’m a backend engineer... ), so you can read it aloud as a script if you want to. Auto-mode toggle + one-shot mic are the two explicit buttons for capture. Auto-mode lets Sona fire on every detected question; one-shot is press-to-listen for a single question. The question filter gates auto-submit so Sona only fires on real questions — not on filler chatter, partial sentences, or your own utterances. If Sona misses a question, switch tab and resubmit manually rather than waiting. For follow-ups on a coding or design answer, use the Sona sidebar on the Coding / Design tabs — it’s a focused Q&A pane that knows the active problem context. Sona is a thinking partner, not a script reader. Use her output as scaffolding — read it, internalize the structure, then answer in your own voice. Recruiters notice when candidates read AI text verbatim, so don’t.

## Hotkeys

Cmd+B (Ctrl+B on Windows / Linux) — instantly blanks the screen. Cariara keeps capturing audio underneath; toggle back when the share ends. Cmd+S — toggle web search assistance on / off. When on, Sona can ground answers with web context. Cmd+Backspace — clear all session history (with confirm dialog).

## Screen-share safety

Cariara runs in a separate browser tab (or desktop window). When the interviewer asks you to share, share your specific window (CoderPad, IDE, doc) — never the full desktop. If you’re forced into a full-screen share, hit Cmd+B immediately — Cariara goes black, audio capture continues, and you can toggle it back when the share ends. The desktop app is not hidden from screen-capture APIs. Treat the web app and desktop app as visually identical for safety: tab-share-only is your discipline either way.

## After the call

Every conversation is saved to /live/sessions . Review the questions you struggled with and feed them into your Prepare notes so the gaps surface in your next study session. You can delete individual sessions or clear the whole history from the sessions page — explicit Delete and Clear-all buttons, nothing hover-only. Cmd+Backspace also clears all sessions globally with a confirm dialog.

## Troubleshooting

Likely cause "No audio detected" stays on after the call starts. Audio Setup Wizard method picked the wrong path, or Chrome share-tab dialog was missing the "Share tab audio" checkbox. Open the wizard, click Disconnect, pick the right method card, click Connect again. Cariara is transcribing your own voice back at you. You’re on room-mic or mic-only capture without voice filtering enabled. Open Settings → Enroll My Voice. Filter enables automatically once enrollment finishes. See Voice filtering in the docs. No answer streams in after a question. Question filter classified the audio as not-a-question. Switch tab and submit manually, or wait for the next clearly-formed question. Sona answered in coding format but the question was design. Auto-routing misclassified. Switch to the Design tab and re-paste the question in the Design input box. Architecture diagram never appears on a design answer. Generated code tripped the sanitizer or the renderer hit a timeout. The prose answer is unaffected. Click Regenerate to bypass the cache and try again. "AI time used up" mid-call. Pool ran out and auto-topup is off. Buy a top-up from Billing, or turn on auto top-up before the call so this never happens. Cmd+B blanks the screen but doesn’t unblank. Focus is in another window. Click the Cariara window first, then press Cmd+B again.

Source: https://docs.cariara.com/live-session