AI-POWERED VIRTUAL MEETING SUMMARIZER

Abstract: Taking notes at meetings manually can be prone to errors, tedious, and often does not clearly delineate actionable follow-up items for team members. This is especially true in remote collaboration settings that inherently do not facilitate real-time insights. In this work, we present a scalable, web-based Virtual AI Meeting Summarizer that provides transcription, summarization, action item extraction, and upcoming proposed sentiment analysis (if requested), in real time. Our tool combines streaming transcription using Deepgram's high-speed ASR API, concise summarization using Large Language Models (LLMs) accelerated by Groq, and action item detection using rule-assisted NLP. The system utilizes a unified Node.js microservices architecture, WebSocket streaming, and browser-native audio capture via a Chrome extension. The architecture allows for secure, low-latency pipelines, PDF export, and persistent storage via a Node.js user service and PostgreSQL. A review of the literature identifies gaps in meeting summarization technologies with respect to unifying these capabilities in real time or privacy-preserving deployments. Our summarizer fills one or a combination of these gaps through a combined, shareable, extendable approach to real-time summarization in organizational and educational contexts.

Keywords: Meeting Summarization, Action Item Extraction, Whisper Model, Transformer Models, Natural Language Processing, Real-Time Transcription.