Turning 50 training videos into a multilingual AI-powered learning platform
BKF Online Schulungen · AI-Powered Driver Training · 2024 – 2025
The Situation
BKF Online Schulungen provides mandatory continuing education for Berufskraftfahrer (commercial truck drivers) in Germany. Drivers complete training modules on safety, compliance, tachograph operation, and equipment handling. Each module includes video content followed by comprehension questions that verify the driver understood the material.
The platform needed to scale its course library to serve drivers across multiple countries and languages. But every new video required a manual, labour-intensive process: someone had to watch the video, write comprehension questions, create correct and incorrect answers, identify the relevant video timestamps for remediation, and repeat the entire process for each supported language.
The Real Problem
Content production was the bottleneck. The platform had strong demand and a clear regulatory need, but the cost and time of producing learning materials for each video, in each language, limited how fast the course catalogue could grow. The team could record new training videos faster than they could produce the accompanying assessments. Every new language multiplied the effort.
This wasn't a technology problem at its core. The videos existed. The knowledge was in the content. The problem was extracting structured learning materials from unstructured video at a cost that scaled.
Key Decisions
- Built an AI-powered content pipeline using OpenAI APIs. When a new video is uploaded, the system automatically transcribes the audio, generating a full text transcript. No manual transcription, no outsourcing to transcription services. The pipeline runs on Kubernetes and integrates with Vimeo for video hosting.
- Automated question generation from transcripts. The AI analyses the transcript and generates comprehension questions along with correct answers and plausible incorrect alternatives. Each question is linked to a specific timestamp in the video, so when a driver answers incorrectly, they're directed back to the exact section that covers the relevant material.
- Multilingual delivery built into the pipeline. Rather than creating content in German and translating afterwards, the pipeline generates questions and feedback in all supported languages as part of the same automated process. Adding a new language is a configuration change, not a content production project.
- Chose Laravel and Livewire for the platform. The training platform runs on Laravel with Livewire for interactive components, providing a straightforward stack that the BKF team could maintain and extend independently. The AI pipeline integrates via backend services, keeping the AI layer separate from the application logic.
The content production bottleneck disappeared. New training videos go from upload to fully assessable, multilingual learning modules without manual content creation. The platform now serves 50+ training videos across 5+ languages with an AI pipeline that handles the entire content lifecycle: transcription, question generation, timestamp mapping, and translation.
More importantly, the economics of the platform changed. Adding new content went from a weeks-long process involving content specialists and translators to an automated pipeline that produces assessment materials in minutes. The team can focus on recording quality training content instead of spending their time on the manual work of turning that content into assessable learning modules.
The Exit
Six-month hands-on engagement spanning 2024 to 2025. I shaped the concept and architecture for the AI-powered content pipeline and built the integration with OpenAI's APIs. The Laravel platform and its ongoing development are owned by the BKF team. The AI pipeline runs autonomously: when new videos are uploaded, the system handles the rest.
I've written about how this project and others shifted my approach to AI in AI Changed How I Work, Not What I Do.
Facing something similar?
Every situation is different but the patterns repeat. If your team is stuck, your architecture is uncertain, or you're about to make a big technical bet, tell me what's going on.