Inspiration

Teams spend hours in meetings, but outcomes often get lost due to manual note-taking, missed decisions, and poor task tracking. We were inspired to bridge the gap between discussions and execution by using AI to make meetings truly actionable.

What it does

Vyimyi is an AI-powered meeting assistant that transforms meetings into actionable outcomes. It joins live meetings or processes recorded audio to deliver real-time intelligence and automation. Key features include: Real-time speech-to-text transcription for live meetings and uploaded recordings AI-powered meeting summaries that capture key discussion points and decisions Automatic action item extraction with clear task ownership Instant conversion of action items into Kanban board tasks, eliminating manual updates Real-time moderation to detect off-topic or unprofessional communication Post-meeting reports containing transcripts, summaries, action items, and insights Flexible usage with support for both live meeting participation and recorded uploads By converting unstructured conversations into structured, trackable tasks, Vyimyi ensures meetings directly translate into execution, accountability, and productivity.

How we built it

Vyimyi was built using a full-stack, AI-driven architecture designed to handle real-time meetings, intelligent processing, and automated task execution end to end. We implemented Firebase Authentication to ensure secure user access with email and Google sign-in. For live meetings, we developed automated meeting bots using Selenium that can join online meetings and capture audio streams from platforms such as Google Meet and Jitsi Meet. This enables the system to participate in meetings without manual intervention. Captured audio—both from live sessions and uploaded recordings—is processed using Google Speech-to-Text, enabling accurate, real-time and post-meeting transcription. The generated transcripts are then analyzed using the Google Gemini API, which performs natural language processing to generate concise meeting summaries, extract key decisions, and identify actionable tasks with assigned ownership. All extracted action items are automatically converted into a Kanban-style task management dashboard, allowing teams to visualize progress, track responsibilities, and move seamlessly from discussion to execution. The entire pipeline integrates real-time audio capture, NLP intelligence, moderation logic, and task automation into a single unified workflow.

Challenges we ran into

  • Capturing and processing live meeting audio reliably
  • Extracting accurate action items and ownership from unstructured conversations
  • Synchronizing real-time transcription with NLP insights
  • Designing seamless Kanban automation without manual intervention
  • Ensuring moderation without disrupting meeting flow

Accomplishments that we're proud of

  • Built an end-to-end AI meeting intelligence platform
  • Achieved real-time transcription and summarization
  • Automatically converted discussions into Kanban tasks
  • Implemented live moderation and post-meeting reporting
  • Delivered a working MVP with dashboards, transcripts, and task boards

What we learned

  • Meetings need execution-focused intelligence, not just recordings
  • Real-time AI processing requires careful orchestration of services
  • NLP accuracy improves significantly when context is preserved
  • Automation drastically reduces follow-ups and accountability gaps
  • End-to-end integration creates far more value than isolated tools

What's next for VyiMyi-AI Meeting Assistant

  • Speaker identification and talk-time analytics
  • Automated meeting bot activation at scheduled times
  • Google Calendar and Outlook integration
  • Email-based summaries and automated follow-ups
  • Deeper analytics for productivity and engagement insights

Built With

  • and-tasks-jwt-authentication-?-secure
  • axios
  • fastapi
  • ffmpeg
  • groq-(whisper-large-v3)
  • jitsi
  • mongodb
  • numpy
  • openai-api
  • pyaudio
  • pydub
  • react-18
  • react-router
  • tailwind-css
  • transcripts
  • uvicorn
  • vite
  • websockets
Share this project:

Updates

Submission history