MeetingMind

Inspiration

In today's fast-paced world, meetings consume a significant portion of our time. Often, we need to be present in discussions that might not require our active participation, but we cannot afford to miss important updates. MeetingMind is designed to help users manage their time efficiently by acting as a virtual meeting agent that participates on their behalf while keeping them informed about crucial moments.

What it does

MeetingMind is an AI-powered meeting agent that transcribes and processes meeting conversations in real time. It simulates participation, generates concise meeting minutes, and determines when a user needs to be notified to "listen in" based on keyword triggers and AI-powered reasoning. If a critical discussion arises, it provides the user with a summary via Discord and allows them to decide whether to join the meeting or continue with their work.

Core Features:

Simulated Meeting Playback: The system processes a transcribed script in real time, simulating the meeting’s natural flow.
AI-Generated Meeting Minutes: A large language model (LLM) generates structured meeting minutes from each spoken line.
Keyword-Based Alerts: A keyword checker scans the dialogue for predefined triggers that might require user attention.
Role-Based Escalation: Another LLM aggregates meeting data over a fixed interval and determines if an unexpected member needs to be looped into the discussion.
Discord Notifications: If user intervention is required, a third LLM fetches relevant meeting context and sends a quick summary to the user, allowing them to decide whether to join.
Agenda and Team Profile Management: Keeps track of meeting agenda items and who is required to participate in each discussion.
Dynamic Engagement Control: If the user joins the meeting, certain AI functions halt until they disengage; once they leave, monitoring resumes.

How we built it

Transcription Handling: The agent processes a simulated meeting transcript with structured timestamps and speaker names.
Time-Based Simulation: Meeting interactions are triggered in real-time based on the timestamps.
Multi-LLM Processing:
- LLM 1 generates meeting minutes from the conversation.
- LLM 2 performs reasoning on aggregated dialogue segments to detect participation needs.
- LLM 3 summarises action items at the end of the meeting.
Keyword Monitoring: A separate function detects predefined keywords that trigger notifications.
User Notification System: Integrated with Discord, allowing the user to receive and respond to alerts in real time.
State Management: Ensures that once a user engages in the meeting, unnecessary notifications are paused, resuming only when they leave.

Challenges we ran into

Real-Time Processing: Coordinating multiple LLM interactions while maintaining seamless meeting simulation was complex.
Defining Context Relevance: Ensuring that the meeting agent correctly identifies key moments that require user attention.
Minimizing False Alarms: Balancing keyword detection and AI-driven reasoning to avoid unnecessary interruptions.
Efficient Agenda Tracking: Dynamically adjusting the need for participation based on the ongoing discussion versus predefined team roles.

Accomplishments that we're proud of

Successfully integrating a multi-step AI workflow that processes meetings efficiently.
Achieving a robust notification system that balances user engagement and autonomy.
Implementing a dynamic participation logic that adapts based on meeting context.

What we learned

The importance of fine-tuning keyword-based triggers to prevent excessive notifications.
How to optimize LLM interactions to balance speed and accuracy in decision-making.
Effective ways to integrate AI-driven reasoning with user-controlled engagement.

What's next for MeetingMind

Enhanced Personalization: Allow users to define custom keyword triggers and meeting preferences.
Multi-Platform Support: Expand notifications to email, Slack, and mobile push notifications.
Advanced Meeting Summarization: Improve LLM capabilities for more structured and insightful meeting minutes.
Integration with Meeting Tools: Support for Zoom, Google Meet, and Microsoft Teams integration.
AI Voice Simulation: Enable AI-generated responses in meetings to interact on the user’s behalf.
More Granular User Control: Provide users with finer control over when and how they are notified during meetings.

MeetingMind is the future of intelligent meeting participation, ensuring users stay informed while optimizing their time efficiently.