AI and Software Engineering
Useful AI, built with foresight.
LLM, RAG, and ML systems shaped from experiment to production with clear interfaces, measured behavior, and production reliability.
AI and software development at SAP. M.Sc. Computer Science at RPTU. Research experience at DFKI across computer vision and privacy-preserving vision-language models.
About
The work sits where research becomes a working system: fine-tuning, retrieval, agents, evaluation, and backend services, with a steady focus on whether the model is useful in the setting it was built for.
That perspective comes from both sides of the stack: research work that demands careful measurement, and production software work where clean interfaces, reliable behavior, and early attention to edge cases matter.
Experience
SAP SE
Building production GenAI and data services for enterprise knowledge operations, including LLM-powered review workflows on SAP Generative AI Hub, documentation quality and governance pipelines, and backend services deployed on SAP BTP.
RPTU Kaiserslautern-Landau
Focused on machine learning, generative AI, software engineering, and reliable AI systems. Deutschlandstipendium scholar.
Thesis: “Differential Privacy in Vision-Language Models: Exploring the Privacy-Utility Trade-off for Technical Drawing Image Captioning.” Researched how differential privacy changes the behavior of a vision-language model for technical figure captioning. The work combined fine-tuning, synthetic data generation, privacy-preserving training, and a custom evaluation suite to look beyond surface-level scores.
DFKI, German Research Center for Artificial Intelligence
Worked on transformer-based object detection for multi-modal medical image and video data, including model development, experimentation, and curation of multi-modal training data.
Built the ViT Token Analysis Toolkit for layer-wise analysis of token importance in Vision Transformers. The project connected model internals, large-scale evaluation, and experiments across efficient ViT variants on HPC.
Cognizant Technology Solutions
Built and maintained production .NET applications, contributed backend features, and improved database and service performance for enterprise software systems.
Self-directed systems for retrieval, agents, evaluation, and enterprise data access.
A RAG service over the EU AI Act with evaluation built into the development loop, so retrieval and answer quality stay visible as the system changes.
A Model Context Protocol server exposing read-only OData v4 services as typed tools, with validation and clean separation between protocol and data access.
An invoice-exception agent that investigates ERP context, cites evidence, and proposes the next action while keeping final posting under human control.
A research-paper assistant that uses a Neo4j knowledge graph for questions where relationships matter as much as semantic similarity.
Skills
