Just some project of which I’m allowed to talk about ;)…


From 2021 to 2022 I led the Fraunhofer Team that took part in the ONTIS project, which aims to develop tools that automate the creation of ground-truth data for training AI methods. In the knowledge-based part of the approach, the core concepts of the application domain and their relationships are formally described in an ontology, which is then populated using data stored in information systems. The goal is to define clear mappings of arbitrary data in an information system to corresponding ontology concepts, thereby providing the missing training examples. The manual inspection and annotation of ontology concepts in databases or file-sharing platforms pose challenges. The application of machine learning methods, such as natural language processing or image classification, can significantly simplify this task. However, the integration of machine learning with knowledge databases is currently being intensively researched by the AI community.


The goal of Climate4Kids is to create an interactive web application tailored for children aged 6 to 12 years old. This application should function seamlessly on both desktop computers and touch devices such as tablets, with screen sizes of up to 10 inches in landscape mode. Anticipated technologies include HTML5, JavaScript, NodeJS, and MariaDB. The client-side component with the user interface will employ HTML5, JavaScript, and PixiJS. The server-side component responsible for user account management and progress tracking will be developed using NodeJS and MariaDB. From 2021 to 2022 I led the Fraunhofer part of the Team of the Consortium.

The web app aims to present the educational content developed by PH Kärnten on the topic of climate change in an engaging manner. This will be achieved through an appealing interface in the form of a drawn landscape image featuring multiple selectable hotspots. Additionally, the user experience will be facilitated by the presence of a mascot that guides users through the web app. Clicking on one of the hotspots triggers an audiovisual picture story that delves deeper into the respective topic for students. For instance, clicking on a car hotspot leads to a picture story about the challenges of future mobility.

Furthermore, users will be able to make learning progress through embedded quizzes and games, with their achievements displayed within the app (referred to as gamification/achievements). The app will also include instructions for experiments that students can replicate either in the classroom or at home to unlock progress within the app.


As part of the “ROMEO” project (Roof Moisture Evaluation), a model for predicting the moisture behavior in flat roofs has been developed. These predictions are based on data from specialized sensors that capture temperature and humidity values. The sensors were previously integrated into the roof structure by our corporate partner/customer. I was a project member from 2021 to 2022. 

Especially in the context of flat roofs, the issue of moisture within the construction often arises, such as leaky roofs, damp insulation materials, and deteriorating beams. To adequately assess the condition of a flat roof, current humidity values often do not suffice. These values are heavily influenced by seasonal conditions and exceptional events. An example of such an exceptional event would be water infiltration due to damage. Even after repair, humidity levels might remain elevated temporarily, even though the roof has been restored to functionality. Rather, a roof’s performance should be measured by its ability to adequately (repeatedly) dry out.

To react promptly to a damaged roof and avoid greater costs, a certain degree of foresight is required. The developed predictive model enables the automation of this complex and time-consuming process.

The approach taken for the predictive model is based on machine learning methods, a subset of artificial intelligence. Preprocessed and structured data are used to train the model. The model learns to predict humidity values based on this training data.


Dooda Books

Dooda is an app for reading aloud picture books for kids. Furthermore, parents can also record their own voice while reading aloud for situations where they are not able to be physically present. For more information checkout One of my tasks at priorIT is to develop and extend dooda for future challenges.



From January to July in 2016 I’ve been working in the temporary project SQUASH (Surgical Quality Assessment in Gynecologic Laparoscopy) where I created a macOS application to easily create, share and extract annotation data from medical experts. Professional surgeons could use the macOS app “SQUASHly” to indicate errors in the operations using a well established error and evaluation system. The gathered data was then later used to train CNN-based classifiers to explore the possibilities for automatic identification of such erroneous parts of an operation.



From 2012 to 2015 I’ve been part in the Next-Generation Video Browsing Project (NGVB).  The goal of the project was to build on our earlier research regarding video browsing and 3D search interfaces and expand it to the domain of mobile devices like smartphones and tablets. “Video browsing” is known as the interactive process of quickly navigating through the content of a video or video collection in an explorative fashion. In difference to video retrieval the purpose of video browsing is manifold: (i) get a quick overview of the content of a video and its structure, (ii) find out where potentially interesting segments are located, and (iii) interactively search for a specific part of content that cannot be found through video retrieval applications. Although many video browsing approaches have been proposed over the past years, there is great potential to significantly improve existing tools by exploiting features of modern mobile devices.