The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Scaling Video AI at RTL with Daan Odijk - #435
Dec 9, 2020
Daan Odijk, Data Science Manager at RTL, shares his expertise in AI and personalized content strategies. He delves into the MLOps journey at RTL, tackling the complexities of ad optimization and video understanding. Daan discusses the intricate challenges of processing large video files and the engineering feats of creating dynamic content selections. The conversation highlights the impact of custom platforms on efficiency and innovation in media, showcasing how AI transforms viewer experiences through tailored recommendations and innovative trailer creations.
AI Snips
Chapters
Transcript
Episode notes
Daan's Background
- Daan Odijk's background is in AI, with a PhD in information retrieval focused on search engine technology.
- He worked at Microsoft Research, analyzing Bing user search behavior and at a Dutch news aggregator startup called Blendl.
RTL's AI Focus Areas
- RTL focuses on three AI application areas: optimization (ad forecasting and allocation), personalization (content recommendations), and content understanding (video analysis).
- Video AI applications include trailer generation, still selection, and mid-roll placement.
Trailer Generation Training Data
- RTL uses existing trailers as positive examples for training trailer generation models but lacks negative feedback.
- They address this by getting feedback from editors on AI-generated trailers and scene selections.