How DPG Media uses Amazon Bedrock and Amazon Transcribe to enhance video metadata with AI-powered pipelines

DPG Media, a leading media company in Benelux, needed a scalable way to extract and generate metadata like summaries, genres, and moods for their growing library of video content. The article describes how they implemented an AI-powered solution using Amazon Bedrock and Amazon Transcribe to automate this process.

Specifically, the article covers:

• The challenges DPG Media faced, including language diversity, varying content volumes, frequent releases, and data aggregation needs.
• The two-step solution: 1) Using Amazon Transcribe to generate accurate transcripts of audio tracks, and 2) Using the Anthropic Claude 3 Sonnet model on Amazon Bedrock to generate detailed metadata from the transcripts.
• The evaluation process for selecting Amazon Transcribe and the Anthropic model, including quality metrics and human validation.
• The benefits of the solution, such as saving days of work generating metadata for a TV series and enabling a richer, more engaging user experience.
• Lessons learned, including the balance between AI and human expertise, and plans to try new models like Anthropic Claude Opus.