Meet Summary

Converts audio from meeting recordings into machine-readable text transcripts, likely using speech-to-text APIs (Whisper, Google Speech-to-Text, or similar) with post-processing to identify speaker boundaries and transitions. The system ingests video/audio files or streams from conferencing platforms and outputs timestamped, speaker-labeled transcript segments that serve as the foundation for downstream summarization and action item extraction.

Processes full transcripts through a large language model (likely GPT-4, Claude, or similar) to generate concise, human-readable summaries that capture key discussion points, decisions, and context. The system likely uses prompt engineering with transcript chunking (to handle long meetings within token limits) and may employ both extractive summarization (pulling key sentences) and abstractive summarization (generating new text) to balance fidelity and brevity.

Parses meeting transcripts and summaries to identify tasks, decisions, and follow-ups using NLP or LLM-based extraction. The system likely uses prompt engineering or fine-tuned models to recognize action item patterns (e.g., 'John will send the report by Friday', 'We need to schedule a follow-up') and structures them as discrete, assignable tasks with implicit or explicit owners, deadlines, and descriptions.

Provides a web interface or API for users to upload meeting recordings (video/audio files) or connect to cloud storage (Google Drive, Dropbox, OneDrive) to retrieve recordings. The system stores uploaded files temporarily or permanently, manages file lifecycle (retention, deletion), and provides access controls for team members. Integration likely uses OAuth for cloud storage and standard file upload APIs.

Provides a user-friendly web interface for viewing generated summaries, action items, and transcripts with search, filtering, and sharing capabilities. The dashboard likely includes a meeting history view, individual meeting detail pages with collapsible sections (summary, action items, transcript), and export options (PDF, email, Slack). Built on standard web frameworks (React, Vue, or similar) with server-side storage and retrieval of processed meeting data.

Implements a freemium pricing tier that allows users to process a limited number of meetings per month (e.g., 5-10 recordings) without payment, with paid tiers unlocking higher limits, team features, and integrations. The system tracks usage per user account, enforces quota limits at processing time, and provides upgrade prompts when limits are approached. Billing likely handled via Stripe or similar payment processor.

Automatically sends processed meeting summaries and action items to attendees via email after transcription and summarization complete. The system likely uses a transactional email service (SendGrid, Mailgun, AWS SES) to deliver templated emails with summary excerpts, action item lists, and links back to the dashboard. Notifications may be configurable per user (digest vs. immediate, opt-in/out).

Automatically extracts and structures meeting metadata (date, time, duration, attendees, title) from recording files, transcripts, or calendar integrations. The system uses filename parsing, audio metadata, or transcript analysis to infer meeting context and organizes meetings chronologically with searchable tags and categories. This metadata serves as the foundation for meeting history, search, and filtering.