Researcher in machine learning,
computer vision and NLP
I am a PhD candidate at Friedrich Schiller University Jena in the Computer Vision Group, where my research centers on fine-grained insect classification with a focus on domain integration. Prior to this, I worked as a machine learning engineer with experience spanning computer vision and large-scale LLM training — from wildlife behaviour analysis using deep learning at Wageningen University, to distributed LLM training at Bittensor, a decentralized AI platform. I have hands-on experience with Python, PyTorch, and Transformers, and have previously contributed to the object detection platform BOX21 and Kaggle classification competitions.
I am currently pursuing a PhD and open to conversations about research engineering, applied ML, and postdoctoral opportunities. Feel free to reach out if my work aligns with your interests.My PhD research focuses on Lepidoptera where I am building a full localization and classification pipeline, investigating dataset composition, model architectures, and auxiliary text information such as taxonomic hierarchies to improve fine-grained recognition. Our group led by Prof. Joachim Denzeler, works on medicinal data, physics, and ecology which stimulates learning across disciplines. I am also a member of ELLIS, the European Laboratory for Learning and Intelligent Systems through which I engage in scientific exchange across European research.
From 2024-2026, I have been contributing to Bittensor, a decentralized network that incentivizes AI development. Living for 2 years as a digital nomad, my programming skills accelerated significantly through active participation in several subnets, including Image Inference and Data Scraping.
I found my niche in distributed LLM training, collaborating with a global team to train a single, large language model. My contributions extended beyond providing GPU compute to direct code contributions to the core repository. This involved deepening my understanding of complex NLP concepts and implementing system improvements, including experimenting with different optimizer implementations and engineering a new, more efficient dataloader.
Prior to this, from 2021–2024, I worked at BOX21 as a Full-Stack Developer, where I gained end-to-end experience deploying object detection models (YOLOv5, MMDetection, DeepLabCut) and building supporting systems with JavaScript, Python, SQL, and Docker.
Kaggle competitor Fathomnet classification competition 2023
Jorrit van Gils, Sean Nachtrab, Lukas Picek
Detecting and classifying marine species in underwater images.
Created shallow vs. deep water animal separator using yolov5
and the Query2Label Transformer.
Wildlife action recognition in camera-trap
photographs using yolov5 and pose estimation
Student: Jorrit van Gils
Advisors: Patrick Jansen, Henjo de Knegt, Helena Russello, Gert Kootstra, Ramon
Holland
[thesis][code/data]
Land-use classification using AlexNet on the UCM satellite dataset
Students: Jorrit van Gils and Lars ter Kate
Applied the AlexNet CNN achitecture and monitored how changes in hyperparameters and
architecture optimised our model. We ended up with a 60% accuracy.
[report]
Computer Vision Courses
Minor Artificial Intelligence at Wageningen University
Course examples:
Teacher Secondary school Biology
From 2016 to 2019, I served as a Biology teacher and mentor at De Nassau High School in
Breda, Netherlands. Seeking to advance my expertise
in data science, AI, and conservation, I pursued a Master's degree in Forest Nature
Conservation at Wageningen University.
Download my CV here.
Since 2020, I have had the privilege of being actively involved with one of the largest nature organizations in the Netherlands, Staatbosbeheer, where I proudly represented the voice of young conservation enthusiasts. As the chair of the communication working group, my responsibilities also included leading meetings, overseeing news updates and managing external communications..
