Yannik Käckenmeister

Building the backend infrastructure for the intelligent document processing and interaction products "IDA" and "Jaide". Responsibilities include designing and implementing interservice communication, caching, data persistence, APIs, authentication and authorization, job lifecycle and workflow management, as well as broader architectural concerns.

Conducted research on sparse indexing structures generated by neural retrievers in RAG systems and designed embedding pruning and expansion methods to improve their efficiency and effectiveness. Concluded with a one-month transition period focused on project handover, while completing my master's degree.

Extended my bachelor’s thesis research on neural 3D object detection by applying a thesis-developed point cloud generator to create synthetic underwater data for model training and evaluation. Co-authored the resulting publication. Conducted a literature review and integrated and evaluated 3D object detection models.

Developed services to ingest, parse and analyze vehicle test-drive and field data, including importers, uploaders, a log parser and a GUI. Also built interactive visualizations of in-vehicle communication and contributed to an app that lets test drivers flag anomalies.

Developed new features for the “4EM” modeling software and supported teaching as a tutor at the Chair of Business Information Systems. Additionally, reviewed students’ preliminary exam work for the Operations Research module at the Chair of Theoretical Computer Science.

Grade: very good | Thesis: "Sparse Index Structures in RAG Systems: Efficient Search and Embedding Strategies for Heterogeneous Data Sets"

Grade: very good | Thesis: "Identification, Evaluation and Adaption of Neural Network Based 3D Object Detection Methods for Underwater Environments"