Working Student (m/f/d) Computer Vision & Machine Learning

Karevo develops AI-powered optical sorting systems for potatoes and onions. Our computer vision models detect defects and foreign objects with high precision, helping farmers ensure quality and reduce losses. You will work in a small, agile team where your ideas have direct impact on the product.

Your Mission

• Improve our existing object detection models through transfer learning and systematic fine-tuning on domain data.
• Evaluate and compare modern model architectures – including Vision Transformers like RF-DETR (Roboflow) – regarding accuracy, latency, and suitability for edge deployment.
• Set up an experiment tracking system (e.g., MLflow) to make training runs, metrics (mAP, Precision, Recall, F1), and model variants comparable.
• Optimize and extend our auto-labeling pipeline to reduce annotation effort for new image data – using CVAT (Computer Vision Annotation Tool) and pre-trained models.
• Structure and manage our image datasets (data preparation, augmentation, train/val/test splits).
• Analyze systematic model failures (False Positives/Negatives), identify causes (data quality, class imbalance, model limitations), and initiate targeted improvements.
• Implement automated evaluation pipelines on validation datasets.
• Improve our partially manual training and deployment pipelines with scripts and CI/CD tools (GitHub Actions, Docker).
• Set up automated workflows that update and evaluate models when new data arrives.

Qualifications

• Studies: Applied Computer Science, Data Science, Machine Learning, Computer Science, or a comparable field.
• Machine Learning: Understanding of supervised learning, model training, and evaluation metrics (mAP, F1, Precision, Recall).
• Computer Vision: Practical experience with object detection models, ideally Ultralytics YOLO; interest or initial experience with transformer-based architectures like RF-DETR.
• DevOps basics: Familiarity with Docker, GitHub/Linux command line.
• Tools: Experience with Python, PyTorch, or TensorFlow; interest in annotation tools like CVAT.
• AI Tools: Familiarity with Claude Code (or comparable AI-assisted development tools) is a plus.
• Personality: Initiative, responsibility, and the ability to identify and solve problems independently.

What to Expect

• Real responsibility: From day 1, you take on tasks with direct impact on our product – no pure assistant jobs.
• Technology at the intersection: AI meets agriculture – you work with edge computing, computer vision, and real physical products.
• Growth: A small team means short decision paths, lots of room for your own ideas, and fast professional development.
• Flexibility: 20 hrs/week, 2 days on-site in Freising