Intelligent Meeting Analytics: Real-Time Insight Extraction from Microsoft Teams Conversations Using Transformer-Based NLP and WebSocket Streaming

The proliferation of virtual collaboration platforms has fundamentally altered how professional teams communicate, coordinate, and make decisions. Despite this shift, a persistent challenge remains: valuable insights generated during meetings—including action items, key decisions, open questions, and discussion threads—are routinely lost or require significant post-meeting effort to retrieve and document. This project presents MeetingMind, an AI-powered real-time meeting analytics extension for Microsoft Teams that addresses this gap by automatically extracting structured intelligence from live audio streams. The system integrates a speech-to-text pipeline based on OpenAI Whisper for audio transcription, followed by a suite of fine-tuned transformer models from HuggingFace for downstream NLP tasks including extractive and abstractive summarization, action item detection with assignee and deadline recognition, intent classification, and topic segmentation. A Python FastAPI backend coordinates real-time data flow using WebSocket connections, ensuring low-latency processing and delivery of insights to a React-based dashboard. The extension is deployed through the Microsoft Teams Toolkit, enabling seamless integration with the existing collaboration environment. Evaluation of the system against benchmark datasets demonstrates strong performance: the summarization module achieves a ROUGE-2 score of 0.41 and ROUGE-L of 0.61, while the action item extraction pipeline achieves an F1-score of 0.78 on a curated meeting transcript dataset. Topic segmentation using a fine-tuned BERT model reaches 84.6% segmentation accuracy. The post-meeting dashboard, tested under simulated meeting loads of up to 90 minutes, processes and renders all extracted insights within an acceptable latency threshold. 
Overall, MeetingMind demonstrates that combining real-time audio processing with transformer-based NLP can substantially reduce the cognitive overhead of meeting 
documentation and improve team productivity.

This work was conducted at Arab International University (AIU), Syria. The official website of the university is: https://www.aiu.edu.sy