Introduction
Hannes Wittek says a ship engine has moods, then immediately corrects himself: not moods, patterns. For decades he heard them before he had words like signal processing. A changed vibration could travel through the body before appearing in a report.
Hannes lives in Hamburg and advises maritime maintenance projects. His entry into AI began with the recognition that sensors can learn what experienced ears have long practiced.
Story of the Path into AI
In a project on motor maintenance, Hannes saw that audio and vibration data could detect patterns, but only if someone labelled the sounds with practical understanding. He was not a programmer. He had to learn to describe experience in a way data teams could use.
He recorded engine sounds, marked unusual noises and learned the basics of signal processing. His first project was a collection of motor recordings commented on by skilled workers, later used for early-warning models. The first model overreacted to a harmless sound after a specific repair. Hannes recognized it immediately.
That case changed the project. Repair history, operating conditions and worker comments were added beside the signal data.
Current Work
Today Hannes trains young engineers to listen before they model. In maritime projects he helps connect machine patterns with maintenance protocols and craft knowledge. When a model flags an anomaly, he asks what happened before the sound: repair, load, weather, cleaning, human intervention.
Systems become more reliable when mechanical experience and algorithmic detection are judged together. Hannes dislikes the idea that AI makes older workers irrelevant. Often they are the ones who can explain why an anomaly is harmless or why a quiet sound is worrying.
Personal Advice
“Sometimes data competence begins with listening carefully,” Hannes says. He advises teams to record not only the sound but also the story around the sound.
Key Facts
Age and place: 68, Hamburg.
Background: retirement, shipyard work, experiential knowledge.
Entry into AI: annotated collection of motor sounds for early detection.
Focus today: maritime industrial AI.
Typical tools: audio sensors, signal processing, maintenance protocols.
Werkstattnotiz
Hannes’s archive includes a recording labelled “harmless after repair, alarming otherwise.” That label is too long for some systems and too important to shorten. He is still testing how context can stay attached to signals after they enter a model pipeline.