Real-time AI meeting copilot built with Next.js + Groq.
This application listens to live microphone audio, converts speech into rolling transcript chunks, and generates 3 intelligent real-time suggestions based on the latest conversation context.
It also supports transcript-aware chat, session export, configurable prompts, and live chunk processing.
- Start / Stop microphone recording
- Automatic chunking every configurable interval
- Final partial chunk processed on stop
- Transcript stored as timestamped chunks
- Auto-scroll to newest chunk
- Combined context window for AI reasoning
Each chunk automatically generates:
- Questions to ask
- Talking points
- Clarifications
- Action items
- Fact checks
Exactly 3 structured suggestions per batch.
Click any suggestion to generate a deeper transcript-aware response.
Manual chat also supported.
- Groq API Key
- Chunk interval
- Prompt tuning
- Token controls
- Temperature
Export full session as JSON including:
- transcript chunks
- suggestions
- chat history
- settings snapshot
- Next.js 16
- React
- TypeScript
- Tailwind CSS
- Groq Whisper Large V3 (speech-to-text)
- Groq LLM for suggestions + chat
- MediaRecorder
- getUserMedia
Mic Input ↓ Audio Chunk Recorder ↓ Groq Whisper Transcription ↓ Transcript Chunks ↓ Suggestion Engine ↓ Chat + Export + Settings
Meetings are noisy and fast.
People miss:
- action items
- blockers
- decisions
- follow-up questions
TwinMind acts like a real-time copilot that listens, summarizes, and suggests what matters most.
npm install
npm run dev