Projects
AI Consulting and Engineering in Public and Financial Sector
As an Associate at PwC Germany, I led and supported various AI-driven projects for clients in the public and financial sectors. My role involved consulting on the implementation and integration of AI technologies to optimize business processes, while also moderating technical discussions with clients. I spearheaded the development of an end-to-end MLOps workflow on Azure Databricks, automating processes such as data preparation, model retraining, and deployment, which ensured the continuous optimization of machine learning model performance.
I also integrated secure, locally operated Large Language Models (LLMs) with Retrieval-Augmented Generation (RAG) support using LangChain, enhancing the quality of AI responses by leveraging local data sources. In addition, I designed and built cloud infrastructures on Azure and developed web applications for data-driven analysis and customer behavior prediction, leading to improved product offerings in the banking sector. My work included creating dynamic dashboards in Power BI and Azure Databricks, conducting comprehensive requirement analyses, developing architectural concepts for Natural Language Processing (NLP) solutions, and implementing algorithms for the automated extraction and processing of unstructured data.
Beyond technical implementations, I designed user-friendly front-end components for an Electron app, optimized LLM-based support functions, and developed internal training materials on cloud-based Generative AI projects. I also crafted proposals and managed client relationships, driving business development and enhancing customer engagement.
AI-Based Surgery Assistance Systems
At the DKFZ German Cancer Research Center, I contributed to advancing AI-based surgery assistance systems through extensive research and development. My work focused on evaluating uncertainty in Neural Networks for surgical image segmentation and conducting in-depth research on Federated Learning algorithms specifically tailored for multi-centric surgical video data. This research aimed to optimize the performance and reliability of AI models in surgical environments by enabling decentralized learning across multiple institutions. The combination of these efforts enhances real-time decision-making capabilities for surgeons, pushing the boundaries of AI’s application in critical medical settings.
Research on Extreme Image Segmentation
As part of my research at the Chair of Machine Learning for Computer Vision at TU Dresden, I explored advanced techniques to improve the robustness of image segmentation algorithms. The project, titled “Extreme Image Segmentation,” focused on understanding the challenges of inaccurate segmentation—specifically, errors where distinct regions are incorrectly merged (False Joins) or where a single region is wrongly split into separate parts (False Cuts). I developed a framework to generate synthetic 3D images that highlight these errors, using computational geometry methods to simulate different error scenarios. This work provides valuable insights into enhancing the accuracy of segmentation algorithms, which is critical in fields such as medical imaging, robotics, and computer vision applications.
Automated Short-Video Creation
This project introduces a fully automated video creation system that allows users to generate high-quality videos with a single click. The system handles the entire process, including narrative development, visual generation, voiceover production, and subtitle synchronization, without the need for manual input.
By automating traditionally labor-intensive tasks such as prompt writing, video editing, subtitle generation, and content sourcing, this solution streamlines video production. The result is custom content that is consistently aligned with HD and 3D visuals, professional voiceovers, and precisely timed subtitles—all generated effortlessly with one click.
Automated Image Creation
Video creation often requires high-quality images to enhance visual content. To address this need, I developed an automated workflow that generates high-resolution images across various topics.
This system efficiently produces visually compelling images suitable for diverse applications. Selected images, inspired by original concepts, have been published on Displate for broader appreciation and use.
