CatchMeUp - Meeting Summariser

Project information

  • Category: Natural Language Processing - Transformers
  • Done for: Internship/University
  • Project date: 2022
  • Project URL: (Not Available due to private repo)

The AI Meeting Assistant project aims to develop an intelligent system to summarize dialogues from online meetings, allowing users to review the content without watching the full recording.

Utilizing an agile, SCRUM-based approach, the team divided tasks into small breakout groups, leveraging diverse skills and regular bi-weekly meetings with sponsors and supervisors for feedback. The primary requirement was a web application capable of uploading and viewing meeting summaries, with options for different summarization methods, customizable keyword focus, and user-friendly UX features like light and dark themes. The project emphasized scalability, ease of use, and security, incorporating AWS for a serverless solution and AWS Cognito for user authentication. Despite evolving requirements, the core functionality remained stable, focusing on creating relevant and informative summaries and ensuring secure, scalable storage for sensitive meeting data. The development involved extensive research into various AI models for natural language processing, ultimately selecting transformer-based models like BERT for extractive summaries and T5 for abstractive summaries. The system uses pre-trained models fine-tuned for summarization tasks, enhanced by libraries like NumPy, PyTorch, SpaCy, and SKlearn for data processing, syntactic analysis, and clustering. The frontend design evolved to include a calendar view and search functionality for better user experience, while the backend used a REST API for maintainability and scalability. The AI models were implemented to provide personalized summaries based on user-input keywords and named entity recognition to highlight specific categories. The project’s iterative approach and frequent SCRUM meetings ensured consistent progress and effective collaboration, positioning the AI Meeting Assistant as a versatile and secure tool for efficient meeting review.