Scribbl

Captures live audio from video conferencing platforms (Zoom, Teams, Google Meet) and converts speech to text with speaker identification, maintaining speaker labels throughout the meeting duration. Uses audio stream interception and real-time speech-to-text APIs with speaker segmentation models to distinguish between multiple participants without requiring manual speaker labeling.

Processes the full meeting transcript through a language model to extract key decisions, action items, and discussion topics, organizing them into a structured summary. Uses abstractive summarization with entity recognition to identify owners, deadlines, and dependencies, then formats output as a hierarchical document with sections for decisions, next steps, and discussion threads.

Scans the meeting transcript and summary to identify commitments, tasks, and action items, then uses NLP to infer owners (by speaker attribution), deadlines (by parsing temporal references), and priority levels. Outputs a structured task list that can be pushed to project management tools via API or webhook integration, with confidence scores for each inferred field.

Stores meeting recordings and transcripts in a centralized, searchable archive with full-text search across transcripts, speaker-specific filtering, and timestamp-based navigation. Uses vector embeddings to enable semantic search ('find all discussions about pricing') and integrates with cloud storage backends (AWS S3, Google Drive, OneDrive) for compliance and retention policies.

Provides native integrations with major video conferencing platforms (Zoom, Microsoft Teams, Google Meet, WebEx) through platform-specific APIs and SDKs, enabling one-click meeting capture without manual setup. Handles platform-specific audio formats, participant metadata, and authentication flows, normalizing all meeting data into a unified schema for downstream processing.

Generates formatted meeting notes documents (Markdown, PDF, Word, HTML) from transcripts and summaries, with customizable templates for different meeting types (standup, 1-on-1, client call, board meeting). Uses template engines to inject meeting data (participants, date, action items, decisions) into pre-designed layouts, enabling one-click export to external tools or email distribution.

Aggregates meeting data across multiple meetings to surface trends and insights: meeting frequency, average duration, participant engagement (speaking time distribution), decision velocity, and action item completion rates. Uses time-series analysis and statistical aggregation to identify patterns (e.g., 'meetings are 30% longer on Fridays') and generates visual dashboards with drill-down capability to individual meetings.

Analyzes meeting transcripts to generate clarifying questions, identify ambiguities, and surface topics that need follow-up discussion. Uses NLP to detect incomplete decisions, conflicting viewpoints, or unresolved questions mentioned during the meeting, then generates suggested follow-up prompts or questions for the next meeting. Integrates with meeting archive to retrieve relevant context from previous discussions on the same topic.

Analyzes speaker diarization data to calculate speaking time per participant, turn-taking patterns, and interruption frequency. Generates visualizations showing who spoke when, identifies dominant speakers vs. quiet participants, and flags potential engagement issues (e.g., one person spoke 70% of the time). Uses statistical analysis to compare against baseline patterns and identify anomalies.

Integrates with calendar systems (Google Calendar, Outlook, iCal) to automatically detect scheduled meetings and enable one-click capture without manual setup. Matches calendar events to video conferencing invites, extracts meeting context (title, attendees, agenda), and pre-populates meeting metadata before transcription begins. Handles recurring meetings and timezone conversions automatically.