Automatic Summarization of 10K Daily Meetings: from Requirements to Production
summary
A team that runs a video-conferencing product creating over 10,000 meeting recordings per day explains how they moved from a one-hour SLA for automatic meeting summarization to a production-ready low-latency pipeline. The talk covers constraints around transcription latency and outlines engineering and operational approaches to speed up generation of summaries so they become useful to users (scaling, pipeline optimization, and deployment choices).
AI-written summary · about
chapters (8)
- 0:00Introduction
- 7:11Problem and solution
- 13:28Speech recognition
- 19:26Summarization architecture
- 25:26Deployment and scaling
- 30:17Results and feedback
- 33:30Key insights
- 34:35Q&A
related talks
38mThe Art of Compression: Crafting Insightful Summaries with LLMs
06mVOICE Summit 2019 | VED Talks | Mohammed AI Tantawy | July 23
30mChallenges & Solutions for AI-Based Message Summarization
04mOne-on-One with Rich Merrett
06mHierarchical Summarization for Longform Spoken Dialog
27mSummarizing & analyzing videos
05mWelcome Remarks
22mConversations With AI: Transforming Conference Learning & Data Engagement