Home Screen - Summary, Transcription, Sentiment Analysis
Multi-language Transcription Page
Email Summary Page
Visualization Page
Live Transcription Page
Test Meeting Template

MeetSmart: AI-Powered Meeting Companion

Inspiration

Meetings are essential for collaboration, but often lack efficient follow-up and documentation. We were inspired to create a tool that automates the tedious parts of meeting workflows—transcription, analysis, summarization, and distribution—so teams can focus on what matters most: making decisions and getting work done. It also allows us to customise the use of tools for example translate the transcription, get specific visualizations based on meeting atendees and sentiments.

What it does

MeetSmart is an AI-driven platform that enhances meeting productivity by:

Transcribing meetings in real-time or from recordings.
Generating concise summaries and action items.
Identifying speakers and analyzing sentiment.
Supporting multiple languages for global teams.
Distributing outcomes via email.
Visualizing key themes and sentiment dynamics.

How we built it

We developed MeetSmart using a modular architecture:

Speech Recognition: OpenAI Whisper and google-translate for accurate transcription.
Backend: FastAPI for integration and service management, Librosa (for audio processing), NLTK (Natural Language Toolkit), NumPy, Scikit-learn
Frontend: React with JavaScript, Framer Motion, and Lucid React for an interactive UI.
NLP & Analysis:
- process_audio.py: Audio-to-text transcription.
- extract_items.py: Regex-based action item extraction.
- summarize_meeting.py: Extractive summarization.
- sentiment_analysis.py: Sentiment scoring with NLTK's VADER.
- email_summary.py: Formats summaries for email distribution.
- realtime_meeting.py: Enables real-time transcription and analysis.
- analyze_meeting.py: Combines modules into a full pipeline.
Visualization: Word clouds and sentiment charts with Matplotlib and WordCloud.

Challenges we ran into

API Integration: We tackled issues like authentication failures and rate limiting by implementing error handling and exponential backoff.
Audio Processing: Dealing with poor audio quality and differentiating speakers in real-time was complex and required tuning.
Semantic Limitations: Our NLP pipeline faced constraints without OpenAI’s higher-tier APIs, which limited some advanced capabilities.

Accomplishments that we're proud of

Successfully built a full-stack meeting assistant integrating transcription, analysis, and visualization.
Implemented SMTP functionality to automatically send transcriptions, summaries, and extracted action items via email to meeting participants.