Browse the meeting like game film. Scrub the video, jump to detected views, inspect stills, and line up transcript spans with the public record.
If a video has embedded captions, extract them into timestamped text spans and preserve the original source track.
When captions are missing, extract mono 16 kHz audio, run local transcription, and align transcript rows to the video timeline.
Every still has a timestamp. Join it with nearby caption spans, OCR text, view labels, and speaker-turn candidates.
Reviewers promote canonical stills for the full board, members, and each public speaker turn.
Pipeline jobs are distributed across a network of workers. Contributors earn bounties for processing video — transcription, OCR, stills, and more.
