Skip to main content

Meetings

Lira's core product — an AI participant that joins your Google Meet calls, listens in real-time, and responds with natural voice when addressed.

How It Works​

1. User pastes a Google Meet link → clicks "Deploy"
2. Backend launches headless Chromium via Playwright
3. Browser joins Google Meet as "Lira AI"
4. Audio Bridge captures all meeting audio (WebRTC → PCM)
5. Audio streams to Amazon Nova Sonic (speech-to-speech)
6. Nova Sonic transcribes, reasons, and generates voice
7. Wake Word Gate checks if Lira was addressed
8. If yes → audio injected back into meeting as Lira's "mic"
9. All participants hear Lira speak naturally

Features​

Real-Time Voice Participation​

Lira doesn't type in chat — she speaks. Using Amazon Nova Sonic's speech-to-speech model, Lira processes audio in real-time and responds with natural, conversational voice including appropriate pauses and tone.

Wake Word Activation​

Lira only responds when someone says her name. The 4-layer wake word detection system ensures:

  • No false positives from similar-sounding words
  • Detection works even when "Lira" is split across audio chunks
  • Configurable cooldown to prevent rapid-fire triggers

Speaker Diarization​

Using Deepgram Nova-2, Lira identifies who is speaking in real-time. Transcripts show real names (e.g., "John: I think we should...") rather than generic labels. This is achieved through:

  1. Deepgram assigns speaker indices from the audio stream
  2. DOM scraping correlates indices with participant names from Google Meet
  3. The correlation improves as more people speak

Meeting Summaries​

After the meeting, Lira generates summaries via GPT-4o-mini:

  • Short summary — 4-6 sentence overview of the key discussion points
  • Detailed summary — 400-700 word breakdown including:
    • Key decisions made
    • Per-person contributions (who said what, who drove the discussion)
    • Action items identified
    • Unresolved questions

Task Extraction​

Lira automatically identifies action items from the conversation and creates structured tasks with:

  • Title and description
  • Assignee (matched to org members)
  • Priority level
  • Integration sync (pushed to Linear, GitHub, etc.)

Voice Commands​

CommandAction
"Lira, mute yourself"Lira acknowledges and goes silent
"Lira, unmute"Lira resumes listening and responding
"Lira, summarize"Provides a real-time summary of the discussion so far
"Lira, create a task for..."Extracts and creates a task

Auto-Leave​

If Lira is the only participant remaining, she automatically leaves after 45 seconds.

Post-Meeting Email​

After the meeting ends, Lira can send email notifications via Resend containing:

  • Meeting summary
  • Task assignments
  • Meeting recording link (if enabled)

Recipients can reply to Lira's emails — she'll read the reply, search the knowledge base, and respond or escalate.

Personality Configuration​

Each meeting deployment can use a different personality mode:

ModeDescription
SupportiveEncouraging, affirming, builds on ideas
ChallengerDevil's advocate, questions assumptions, pushes for rigor
FacilitatorKeeps discussion on track, ensures all voices are heard
AnalystData-driven, focuses on metrics, evidence, and outcomes

Demo Meeting​

For testing without Google Meet, use the Browser-based Demo Meeting:

  1. Open the demo from your dashboard
  2. Grant microphone access
  3. Talk to Lira directly via WebSocket
  4. Full AI experience without a Google Meet link