What does an AI scientist actually do in a microbiology institute?

Invited talk for the Master program Applied Computational Life Science at ZHAW.

In this talk, I gave a practical view of what an AI scientist actually does inside a microbiology institute. While machine learning often sounds glamorous, much of the real work is hands-on: collecting data from different sources, cleaning it, harmonising formats, and making it usable for analysis. At the Institute of Medical Microbiology, this means working with very different kinds of data generated in routine diagnostics and research, including MALDI-TOF mass spectrometry, antibiotic susceptibility testing images, whole-genome sequencing, and large metadata tables.

I explained how different machine learning approaches fit different data types and problems. In some cases, deep learning and foundation models are the right tools, especially for omics and imaging data. In others, simpler approaches such as random forests or gradient boosting remain the most robust and practical solutions. I also described my broader role in harmonising data, developing and deploying code, and coordinating machine learning projects across the institute. Finally, I presented concrete examples from our work, including antimicrobial resistance prediction and outbreak detection from MALDI-TOF data, inhibition zone measurement from plate images, and resistance prediction from genomic data.