AI Meeting Assistant: A Multilingual System for Video Transcription, Summarization, and Context-Aware Question Answering
Online meetings, recorded lectures, and educational videos often contain useful information, but going through long recordings to find specific details can take considerable time. This becomes even more challenging when audio contains different languages or when users need to identify important decisions and follow-up tasks from a discussion. This paper presents an Artificial Intelligence (AI)-based Meeting Assistant designed to make recorded content easier to understand and review. The system accepts YouTube video links and locally uploaded audio or video files. It extracts and processes audio, converts speech into text using speech recognition models, and generates meaningful summaries with the help of a large language model. It also identifies important meeting information, such as action items, decisions, and unanswered questions. The system uses Whisper and Sarvam AI as transcription options, while Mistral AI is used for summarization and information extraction. SQLite stores meeting records and related information, whereas ChromaDB supports semantic search over transcripts. A Retrieval-Augmented Generation (RAG) pipeline allows users to ask questions about a previously processed meeting and receive answers based on its transcript. The proposed system brings transcription, summarization, structured information extraction, and question answering together in one application. It aims to reduce the effort required to review long recordings and make important meeting information easier to access.
Authors
- Himanshi Suman
- Deepti Pant
- Aditi Singh
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-15
- DOI
- https://doi.org/10.5281/zenodo.24256271
- Primary Topic
- Speech Recognition and Synthesis
- Type
- article
- Field-Weighted Citation Impact
- 0.00