AI-Powered Meeting Assistant: An LLM-Centric, Agentic AI Approach for Automating Post-Meeting Workflows

Title

AI-Powered Meeting Assistant: An LLM-Centric, Agentic AI Approach for Automating Post-Meeting Workflows

Authors

1. Nikit Patel, GLS University, Ahmedabad, Postdoctoral Researcher, India
2. Kaushal Patel, System Level Solutions: SLS, Anand, Developer, India

Abstract

Meetings are critical for collaborative decision-making, yet their outcomes are often underutilized due to inefficiencies in capturing, organizing, and tracking discussions. Traditional approaches to meeting documentation—manual note-taking or transcription-based solutions—fail to provide actionable insights, frequently leading to lost information, missed deadlines, and lack of accountability. With the advent of Large Language Models (LLMs) and agentic artificial intelligence (AI), it is now possible to design systems that not only document but also act upon meeting outcomes. This paper presents an LLM-centric AI-powered meeting assistant architecture that automates the complete post-meeting workflow. The system ingests audio or text transcripts, preprocesses them for clarity, and leverages LLMs to generate structured meeting summaries, extract actionable tasks, and create formal Minutes of Meeting (MoM). These outputs are converted into JSON format for seamless integration with task trackers, notification systems, and centralized dashboards. By adopting an agentic AI approach, the system enables proactive follow-ups, real-time reporting, and task completion monitoring through both tracker APIs and email-based confirmations. We provide a detailed literature review of speech-to-text technologies, LLM-driven meeting automation, and workflow orchestration, followed by a comprehensive description of the system’s architecture and methodology. The benefits, limitations, and challenges—including speaker diarization, overlapping speech, and task misclassification—are critically examined. The study highlights the transformative potential of agentic AI for enterprise productivity, while emphasizing ethical considerations and the importance of human oversight.

Keywords

Meeting Automation Large Language Models Agentic AI Speech-to-Text Action Item Extraction Task Tracking Minutes of Meeting Workflow Orchestration Enterprise Productivity

PDF

This browser does not support PDFs. Please download the PDF to view it: View the PDF.

Conclusion

This paper has presented a comprehensive LLM-centric architecture for an AI-powered meeting assistant, designed around the principles of agentic AI. By unifying speech recognition, LLM-driven summarization, structured JSON conversion, and workflow integration, the system automates the entire post-meeting lifecycle. The literature review highlighted advances in ASR, meeting summarization, task extraction, and workflow orchestration that underpin the design. The methodology demonstrates how speech inputs can be transformed into actionable outcomes seamlessly integrated with task trackers, notifications, and dashboards. While significant benefits are evident, challenges such as speech variability, ambiguous task phrasing, and ethical considerations remain. Addressing these will require ongoing research in agentic AI, multimodal systems, and human-AI collaboration. Nevertheless, the proposed system represents a major step toward transforming meetings from passive discussions into actionable, accountable workflows, unlocking new levels of organizational productivity.

Reference

1. -

Author Contribution

The sole responsibility for the study design, data gathering, results analysis, and manuscript drafting lies with the author.

Funding

This work did not receive any specific grant from funding agencies in the public, commercial, or non-profit sectors for its research, authorship, or publication.

Software Information

Not applicable.

Conflict of Interest

All authors declare the absence of any conflicts of interest.

Acknowledge

I extend my gratitude to everyone who contributed their expertise to this study and manuscript, and to the anonymous reviewers for their helpful comments.

Data availability

Not applicable.