Postdoctoral researcher · Computer vision for biodiversity monitoring
UFZ & iDiv · Leipzig, Germany
I am passionate about building software that helps ecologists see more of nature with less effort.
From 2016 to 2025 I worked at the Helmholtz Centre for Environmental Research (UFZ) and the German Centre for Integrative Biodiversity Research (iDiv), first as an R programmer and then as a data scientist, on plant–pollinator networks, statistics, software tools and R packages for ecological analysis. Alongside this job I completed my PhD, combining computer vision, machine learning and affordable smartphones to monitor pollinators automatically. Since July 2026 I have continued this work at UFZ and iDiv as a postdoc.
I am open to research and data science roles in academia and industry. The quickest way to reach me is through LinkedIn. Email works too (valentin.stefan at idiv.de or ufz.de), but my inbox gets a lot of spam, so LinkedIn is faster.
Smartphones are affordable, robust and everywhere. Why not turn them into field cameras that record and recognise insects and other animals? Monitoring biodiversity would then no longer need expensive equipment, technicians to assemble microcomputers, or many hours of sorting images by hand.
An AI wildlife camera on your phone
FaunaPulse turns an Android smartphone into a camera that detects, tracks and identifies animals.
A second life for old phones as time-lapse cameras
FaunaLapse, the sister app of FaunaPulse, turns old or inexpensive Android phones into time-lapse cameras. It needs no AI, so it runs even on phones from as far back as 2017.
Code and data for our 2025 Scientific Reports paper testing pretrained YOLO detectors on unseen time-lapse images of pollinators.
Tools to convert and quantise detection and identification models so they run on phones inside FaunaPulse.
Whittaker biome diagrams with ggplot2, for placing study sites in climate space.
Geodesic buffers around points, with radii in metres.
Bootstrapping indices of bipartite ecological networks, such as plant–pollinator networks.
Turnover saturation curves via bootstrapping, to check whether sampling captured community turnover.
The full list is on Google Scholar and ORCID.
Over the years I have also written R tutorials and plotting scripts. They are collected under Scripts & graph gallery. Older posts live on my WordPress blog, and I have answered many R questions on Stack Overflow.