An AI pipeline that tags, categorises, and summarises a video library at scale.

Computer vision and speech-to-text feeding an LLM summariser — turning a 200k-hour video library into a searchable, structured archive.

200k+hours of video processed end-to-end

At a glance

The challenge, approach & outcome.

Challenge
A digital media platform's 200,000-hour video library was almost entirely unsearchable. Editors hand-tagged the top 5%; the long tail rotted. Search-driven engagement was capped because the catalogue was dark.
Approach
Built a pipeline that ingests video, runs scene detection and computer vision tagging, transcribes audio via Deepgram, and feeds the transcript + visual tags into an LLM summariser. Outputs structured metadata: scene tags, speakers, topics, summaries, and chapter markers — all indexed for search and recommendation.
Outcome
200,000+ hours processed end-to-end in eight weeks. Average processing time of 9 minutes per hour of video. Search-driven engagement up 47% because the long tail of the catalogue is now discoverable.

Deliverables

Stack

Have a similar problem?

30-minute discovery call. No NDA gatekeeping. We'll tell you straight whether we're a fit.