Introduction
Pavel Kranz still hears the forklift before he remembers the image. It came too close around a blind corner, the kind of near accident that leaves everyone loud for ten minutes and quiet afterwards.
Pavel lives in Vienna and works on computer-vision systems for construction and warehouse safety. He entered the field through dust, back pain and the practical knowledge of where danger gathers.
Story of the Path into AI
After the near miss, Pavel asked whether cameras could recognize hazardous situations earlier. He did not want workplace surveillance disguised as safety. He wanted prevention that understood real building sites: rain, dust, poor light, hidden views, tools left where no plan says they should be.
He had no formal IT degree and spoke first in examples rather than technical terms. Those examples became his strength. He learned image annotation, safety standards and the limits of camera models. His first dataset contained messy construction scenes with reflective vests, blocked sightlines and uneven paths.
The first model worked well in tidy test images and poorly where a site actually looked like work. A shadow became a person; a person behind material disappeared. Pavel pushed for more realistic training data.
Current Work
Today Pavel works with a safety-tech company. He reviews model outputs and argues for warnings that help crews rather than punish them. In one pilot, a model treated open areas as safe. Pavel knew from experience that open areas often fill with material at short notice. The rules were changed.
Pilot customers now receive fewer false alarms because the system is trained against real site logic. Pavel remains wary. A camera can warn, but it cannot carry responsibility for planning, staffing or a culture where people feel allowed to stop unsafe work.
Personal Advice
“Anyone automating safety should first know how danger smells and sounds,” Pavel says. He tells technical teams to walk the site before labelling the data. The dangerous thing is rarely the object alone; it is the situation around it.
Key Facts
Age and place: 41, Vienna.
Background: physical labour, health pressure, career change.
Entry into AI: dataset with realistic construction-site safety scenes.
Focus today: occupational safety and computer vision.
Typical tools: computer vision, annotation, safety standards.
Werkstattnotiz
Pavel’s favourite training image is ugly: rain on the lens, a half-hidden pallet, two workers crossing paths. It lowers benchmark scores and raises trust. He keeps asking whether the next dataset contains enough bad weather to be honest.