Made by Brian Hui, Eric Kwon and Abbe Azale

An intelligent meeting assistant that automatically detects calls, transcribes conversations in real-time, and generates AI-powered summaries, meeting minutes, and action items.

FocusNote is a desktop application that monitors your computer for active calls on Discord, Zoom, or Microsoft Teams. When a call is detected, it automatically:

• Records system audio and microphone input
• Transcribes speech in real-time using Whisper
• Generates meeting summaries, formal minutes, and action items using Gemini AI
• Saves all outputs organized by date and time

• Automatic Call Detection: Monitors Discord, Zoom, and Teams for active calls
• Real-time Transcription: Uses Whisper AI for accurate speech-to-text
• Smart Audio Capture: Records both system audio and microphone on macOS and Windows
• AI-Powered Analysis:
  • Concise meeting summaries
  • Formal meeting minutes
  • Actionable items with context
• User-Friendly UI: Clean PyQt6 interface with live status updates
• Organized Output: All transcripts and AI outputs saved with timestamps

FocusNote consists of three main components:

1. Desktop App (DesktopApp/): PyQt6 GUI application that handles call detection and audio recording
2. Transcription Server (DesktopApp/src/transcription/): Whisper-based real-time speech-to-text service
3. Meeting Microservice (MeetingAssistant/): Gemini AI service for generating summaries and action items

• Python: 3.11 or higher
• Operating System: macOS or Windows
• ffmpeg: Required for macOS audio capture

  # macOS
  brew install ffmpeg
  
  # Windows
  # Download from https://ffmpeg.org/download.html

• Gemini API Key: Required for AI features
  • Get one at Google AI Studio

We provide startup scripts that automatically install/update dependencies and launch all three components in separate terminal windows.

cd DesktopApp
bash scripts/start-all.sh

Or make it executable first:

chmod +x scripts/start-all.sh
./scripts/start-all.sh

cd DesktopApp
scripts\start-all.bat

The script will:

1. Check and install/update all Python dependencies
2. Open three terminal windows:
   • Transcription Server - Whisper AI (port 17483)
   • Meeting Microservice - Gemini AI (port 8888)
   • Desktop App - FocusNote UI

Note: The first run may take a few moments to install dependencies. Subsequent runs will be faster as pip only updates changed packages.

git clone [email protected]:kyulyeon/focusnote.git
cd focusnote

Create a .env file in the MeetingAssistant directory:

GEMINI_API_KEY=your_api_key_here
PORT=8888

Important: Never commit your .env file or API key to version control!

The startup script automatically installs dependencies, but if you prefer to install them manually:

Desktop App:

cd DesktopApp
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

Meeting Microservice:

cd MeetingAssistant
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

If you prefer to start components individually:

Terminal 1 - Transcription Server:

cd DesktopApp/src/transcription
python server.py

Terminal 2 - Meeting Microservice:

cd MeetingAssistant
python meeting_microservice.py

Terminal 3 - Desktop App:

cd DesktopApp/src
python main.py

1. Start the application using one of the methods above
2. Join a call on Discord, Zoom, or Teams
3. FocusNote automatically detects the call and starts recording
4. View live transcription in the console
5. When the call ends, FocusNote automatically:
   • Stops recording
   • Sends transcript to AI service
   • Generates summary, minutes, and action items
   • Saves all outputs to DesktopApp/meeting_output/

All meeting data is saved in DesktopApp/meeting_output/ organized by timestamp:

DesktopApp/meeting_output/
├── 2025-11-09T14:30:45/
│   ├── meeting_summary.txt      # AI-generated summary
│   ├── action_items.txt         # Extracted action items
│   └── meeting_minutes.txt      # Formal meeting minutes
└── meeting_recordings/
    └── meeting_discord_20251109_143045.wav

focusnote/
├── DesktopApp/                     # Main desktop application
│   ├── src/
│   │   ├── main.py                 # Application entry point
│   │   ├── ui/                     # PyQt6 user interface
│   │   ├── audio/                  # Audio capture logic
│   │   ├── detection/              # Call detection (Discord, Zoom, Teams)
│   │   ├── transcription/          # Whisper transcription server
│   │   │   ├── server.py           # Transcription WebSocket server
│   │   │   └── websocket_client.py # Client for real-time transcription
│   │   └── api/                    # Microservice communication
│   ├── scripts/
│   │   ├── start-all.sh            # macOS/Linux startup script
│   │   └── start-all.bat           # Windows startup script
│   ├── meeting_output/             # AI-generated outputs (created automatically)
│   ├── meeting_recordings/         # Audio recordings (created automatically)
│   ├── requirements.txt            # Python dependencies
│   └── README.md
│
├── MeetingAssistant/               # AI microservice
│   ├── meeting_microservice.py     # FastAPI service
│   ├── test_service.py             # Test script
│   ├── requirements.txt            # Python dependencies
│   └── README.md
│
└── README.md                       # This file

The Meeting Microservice exposes the following endpoints:

POST http://localhost:8888/summary

POST http://localhost:8888/minutes

POST http://localhost:8888/action-items

GET http://localhost:8888/health

Request format:

{
  "transcript": "Meeting transcript text...",
  "meeting_title": "Optional title",
  "meeting_date": "Optional date",
  "participants": ["Optional", "list"]
}

Desktop App:

cd DesktopApp
pip install -r requirements-dev.txt
pytest

Meeting Microservice:

cd MeetingAssistant
python test_service.py

cd DesktopApp
python src/detection/detect_test.py --test

• Ensure .env file exists in MeetingAssistant/
• Verify your API key is correct
• Restart the microservice after creating/updating .env

• Port 8888 or 17483 is being used
• Change PORT in .env or kill the conflicting process

• This has been fixed in the latest version
• Sample rates are now properly matched (48kHz)
• Mono mic audio is converted to stereo for mixing

• Ensure Discord/Zoom/Teams is actually in a call
• Check CPU usage is above the threshold (actively transmitting audio)
• Wait for 3 consecutive detections (3 seconds)

• Install ffmpeg: brew install ffmpeg
• Ensure microphone permissions are granted in System Preferences

• Verify all three components are running
• Check the transcription server is on port 17483
• Check the meeting microservice is on port 8888
• Verify internet connection for Gemini API

• Uses ffmpeg for system audio capture
• Requires microphone permissions
• Audio is captured at 48kHz stereo

• Uses PyAudioWPatch for loopback audio
• May require running with administrator privileges
• Supports WASAPI loopback

• All processing happens locally except AI generation
• Audio recordings stay on your machine
• Only transcripts are sent to Gemini API
• API keys are stored in .env files (git-ignored)
• No data is collected or transmitted to third parties

• Python 3.11+
• PyQt6
• PyAudio (Windows: PyAudioWPatch)
• Whisper (pywhispercpp)
• FastAPI
• Google Generative AI SDK
• ffmpeg (macOS)