Parrot AI

Parrot AI captures live video conference streams from Zoom, Microsoft Teams, and Google Meet through native integrations or browser-based recording, applies automatic speech recognition (ASR) to convert audio to text, and performs speaker diarization to attribute transcript segments to individual participants. The system likely uses a combination of platform APIs for direct stream access and fallback audio extraction from recorded files, with ASR processing happening either in real-time or post-conference depending on platform constraints.

Parrot AI indexes transcribed meeting content using a full-text search engine (likely Elasticsearch or similar), enabling users to query across all historical meetings by keyword, phrase, or speaker name. The search likely supports boolean operators and phrase matching, with results ranked by relevance and timestamp. Indexing happens asynchronously after transcription completes, making transcripts searchable within minutes of meeting end.

Parrot AI enables users to share individual meeting transcripts or collections of transcripts with other users via shareable links or direct invitations. The system likely implements role-based access control (viewer, editor, admin) and generates unique tokens for share links, with optional expiration dates and password protection. Shared transcripts can be viewed in a web interface without requiring the recipient to have a Parrot AI account, reducing friction for external stakeholder access.

Parrot AI integrates with Zoom, Microsoft Teams, and Google Meet through native APIs or browser extensions, automatically detecting when a user joins a supported conference and initiating recording/transcription without manual intervention. The integration likely uses OAuth for authentication and platform-specific webhooks or polling to detect meeting state changes. Browser extension approach enables capture from any web-based conferencing tool, while native integrations provide deeper access to audio streams and metadata.

Parrot AI stores transcribed meetings in a searchable database with associated metadata (meeting date, duration, platform, attendees if available, calendar title if integrated). Transcripts are organized chronologically or by custom folders/tags, with quick access to recent meetings. The storage backend likely uses cloud object storage (S3-like) for transcript files and a relational or document database for metadata and indexing, with automatic cleanup policies for deleted meetings.

Parrot AI enables users to export meeting transcripts in multiple formats (plain text, PDF, Word, SRT subtitle format) for use in external tools or archival. Export likely includes options to include/exclude timestamps, speaker labels, and metadata. The export process is likely asynchronous for large transcripts, with downloads available via email or direct link.

Parrot AI captures video recordings of meetings (in addition to audio transcription) and stores them alongside transcripts, enabling users to playback the recording with synchronized transcript highlighting. Clicking a transcript line jumps to the corresponding timestamp in the video, and vice versa. Recording is stored in cloud object storage with streaming playback, likely with adaptive bitrate encoding for bandwidth optimization.

Parrot AI provides a web-based transcript viewer with keyboard shortcuts for navigation (jump to next/previous speaker, search, timestamp jump), accessibility features (high contrast mode, text size adjustment, screen reader support), and responsive design for mobile/tablet viewing. The viewer likely uses semantic HTML and ARIA labels for accessibility compliance, with transcript text selectable for copying.

Parrot AI tracks meeting attendees and identifies speakers in transcripts by integrating with conferencing platform APIs to retrieve attendee lists and speaker metadata. Speaker identification is performed through a combination of audio fingerprinting, voice recognition (if enabled), and platform-provided speaker labels. The system associates speaker names with transcript segments, enabling speaker-specific search and filtering.