RTLZWEI Hones Competitive Edge with Red Hat OpenShift AI

Overview

RTLZWEI, a German broadcaster and digital media company, modernized its on-premise core system BOBY by deploying Red Hat OpenShift Platform Plus and Red Hat OpenShift AI, with guidance from Red Hat Consulting and Red Hat Training. A proof of concept integrated NVIDIA H100 GPUs with Red Hat OpenShift AI, using Whisper on vLLM to speed up video transcription for subtitles, dubbing and accessibility. By fine-tuning the transcription model with its own data, RTLZWEI reduced the word error rate by 33%. The platform also hosts LLMs for data science experiments, forecasting and fine-tuning, running on-premise to meet data sovereignty and cybersecurity requirements.

The challenge

RTLZWEI needed to modernize its core multifunctional on-premise system, BOBY, and adopt a standard platform and framework to speed up AI developments. The company also wanted a platform that provided the freedom and flexibility to experiment with AI within a defined framework, while instilling governance and best practices over AI development, and lacked clearly defined roles between developers and platform engineering teams, which caused confusion and slowed issue resolution.

The solution

RTLZWEI partnered with Red Hat to implement Red Hat OpenShift Platform Plus to accelerate application development and deployment, then rolled out Red Hat OpenShift AI and worked with Red Hat Consulting to implement its use cases, using Red Hat Training to develop in-house AI skills. Its most significant proof of concept integrated NVIDIA H100 GPUs with Red Hat OpenShift AI, using Whisper on vLLM to speed up video transcription for subtitles, dubbing and accessibility, and fine-tuning the transcription model with its own data. The platform is deployed on premise for data sovereignty, though it also runs on self- and fully managed AWS due to its hybrid nature, and hosts LLMs for speech-to-text, data science experiments, fine-tuning and data exploration to support forecasting.

Reported business value

By fine-tuning the transcription model with its own data, RTLZWEI reduced the word error rate by 33%. The company also reports modernized application deployment and development, consolidated IT sprawl into one unified platform, accelerated sovereign AI development with in-house skills, increased competitive advantage in a rapidly evolving media landscape, and an improved IT-security posture.