Berlin, Berlin, Germany
I’m a postdoctoral researcher turned product-minded ML engineer focused on generative AI, LLMs and agentic systems. I translate cutting-edge research (transformers, VLMs, generative audio) into robust, production-ready solutions and enjoy leading teams from prototype to deployed product. I build generative and perception systems that move beyond papers into products. During my PhD (in collaboration with Honda Research Institute EU) I developed meta-ML methods for robot self-assessment in human-robot teaching scenarios — a practical lesson in reliability and real-world integration. In my postdoc at the National Institute of Informatics (NII) in Tokyo I extended this work to generative AI: training 2D latent audio maps with VAEs to condition transformer-based audio generation and combining object detection (YOLO-style) with multimodal LLMs (GPT-4V) to create flexible perception frameworks for autonomous drones. My strengths lie at the intersection of deep technical expertise and pragmatic product thinking: selecting the right models, building robust evaluation and explainability pipelines, and shipping reproducible systems. I’m especially interested in agentic interfaces and LLM-driven automation that augment human workflows rather than replace them. Open to CTO/co-founder or technical lead roles where I can help turn generative AI research into impactful products.
- Postdoctoral Fellow of DAAD (German Academic Exchange Service) Project: Vision Architectures for Drones (4 first-author publications) - Investigated single-stage Object Detection approaches, such as YOLO variants, using Python and Torch. - Introduced Explainable AI approaches for visualizing YOLO gradients, learned features, and salience maps. - Investigated LLMs/VLMs for zero-shot classification in aerial images using OpenAI API and locally deployed models. - Developed generative models in Torch and Torchvision/Torchaudio for robot control using Reinforcement Learning in realistic simulation environments and for conditioned audio synthesis using Transformer and intelligent user-interfaces.
- Developed models for enabling the applicability of robots in daily living environments and human-robot teaching scenarios. - Implemented service robot demos, including perception and navigation with human-in-the-loop in ROS, Python, C++.
- PhD Project: Competence-based human-machine interaction models for cooperative intelligence in task completion. - Contributed to advancing the institute's mission of Cooperative Intelligence. Published Python libraries for human-aware active and incremental Machine Learning using Tensorflow, Keras, and Torch. - Developed an intelligent robot user interface with JavaScript web frontend and backend in Python, Flask, and PostgreSQL.
- Project: Electronic Environment Recognition with Combine Harvesters. - Master Thesis: Vision-based anomaly detection in grain fields. - Developed image segmentation and single-class classification approaches for anomaly detection using Matlab. - Acquired training data within harvesting field tests.