MeetingMind Live

Your silent AI co-pilot for every meeting - before, during, and after.

Comment

Inspiration

Every meeting leaves a trail of missed facts, forgotten action items, and follow-up emails that never get sent. I wanted to build something that works invisibly — not another notetaking app you have to babysit, but a true AI agent that has your back before you even say hello.

What it does

MeetingMind Live is a 3-phase AI meeting co-pilot powered by Google Gemini:

Phase 1 — Pre-Meeting Briefing: Reads your calendar, past notes, and documents from Cloud Storage. Delivers a spoken briefing and flags what to watch out for. Phase 2 — Battlefield Mode: Runs 4 silent AI layers during the meeting — a Fact Shield that catches wrong numbers in real time, a Question Anticipator that surfaces answers before you're asked, a Negotiation Assistant that holds your deal limits, and a Silent Task Logger that queues verbal assignments for one-tap confirm. Phase 3 — Post-Meeting Executor: Fully autonomous. Generates the summary, extracts action items with owners and deadlines, drafts follow-up emails per attendee, and saves everything to Cloud Storage — before you close your laptop.

How we built it

Backend: FastAPI + Google ADK orchestrating 3 agents on Cloud Run AI: Gemini 2.0 Flash Live API for real-time audio transcription and diarization, Gemini 2.0 Flash for vision analysis and summarization Storage: Firestore for live session state, Cloud Storage for documents and post-meeting summaries Frontend: Single-page web app with WebSocket connection, live transcript feed, and real-time insight cards Infrastructure: Cloud Run with Secret Manager, auto-scaling, and persistent WebSocket sessions

Challenges we ran into

The hardest part was the real-time audio pipeline — getting continuous WebSocket audio chunks into the Gemini Live API with the correct format, silence detection, and low enough latency to feel instant. Orchestrating 3 phases as a stateful agent — where Phase 3 only triggers after Phase 2 completes cleanly — required careful async design to avoid silent failures dropping the entire post-meeting flow. Accomplishments that we're proud of

A fully autonomous 3-phase agent that requires zero manual input after the meeting starts Real-time fact checking and negotiation assistance with sub-second response times Phase 3 that executes entirely on its own — summary, action items, emails, and Cloud Storage save — without the user lifting a finger A clean, minimal UI that stays out of the way and only surfaces what matters

What we learned

Building with the Gemini Live API showed how powerful real-time multimodal agents become when latency is low enough to feel human. The line between "tool" and "agent" disappears when the system can perceive, reason, and act faster than a human notices. Stateful multi-phase orchestration is also much harder than it looks — every phase transition is a potential failure point. What's next for MeetingMind Live

Full speaker diarization to identify who said what

Google Calendar integration to auto-populate meeting context before you even open the app Chrome extension to overlay directly on Google Meet without switching tabs Mobile companion app for on-the-go briefings and voice summaries

Built With

Share this project:

Updates

Submission history