Apurva AI
Real-Time AI Meeting Assistant
Overview
Real-time AI meeting assistant using OpenAI GPT and WebSocket connections for live transcription and enterprise insights.
Tech Stack
Key Highlights
Implemented real-time audio transcription pipeline with sub-2-second latency
Built GPT-powered meeting summarization with action item extraction
Designed WebSocket architecture handling 500+ concurrent meeting sessions
Created enterprise dashboard with meeting analytics and team productivity insights
The Challenge
Enterprises needed an AI-powered tool that could join meetings in real-time, transcribe conversations accurately, and generate actionable summaries with key decisions and follow-ups. The system needed to handle multiple concurrent meetings with minimal latency.
My Approach
I built the real-time pipeline using WebSocket connections for audio streaming, feeding into a speech-to-text engine with speaker diarization. The transcription stream is processed by GPT models to generate live highlights during the meeting and comprehensive summaries afterward. The architecture uses message queues to handle transcript processing asynchronously, keeping the live experience responsive.
Key Contributions
Developed the WebSocket server infrastructure that manages concurrent meeting sessions with automatic reconnection and state recovery. Built the GPT integration layer that produces structured summaries with tagged action items, decisions, and questions. Created the frontend meeting viewer with synchronized transcript playback, searchable highlights, and exportable reports.