AI-Powered Meeting Summarization with Task Extraction and Smart Scheduling

Meetings are where teams make decisions and assign work, yet the knowledge they produce is captured poorly: manual minutes are incomplete, action items are scattered across notes, e-mail and chat, deadlines are forgotten, and a decision taken in one meeting may be silently contradicted in another. This paper presents MeetX, an AI-powered meeting intelligence platform that converts spoken meeting content into structured, verifiable and actionable information. Audio is captured from a Google Meet bot integration or the browser microphone and transcribed with a Whisper-based component. Llama 3 models served by Groq, orchestrated by a Python FastAPI service with LangChain, generate executive summaries, key takeaways, decisions and action items, each task carrying an assignee, due date, priority and confidence score. Low-confidence tasks enter a human-in-the-loop review queue, and approved tasks are dispatched through Gmail, Slack webhooks and Google Calendar. Distinctive modules include a SHA-256 decision ledger with Markdown ADR export, decision-reversal tracking with a proposed Team Decision Stability Index, DAG-based cross-meeting deadline-conflict analysis, a retrieval-augmented “Ask AI” memory and bi-directional status synchronization. The front end uses React 18 and TypeScript; the back end uses Spring Boot 3 with PostgreSQL. This paper reports the design, mathematical model and evaluation plan of the system. It is a preliminary work, and no measured results are claimed.