AI Engineer & Data Scientist | LLM · NLP · Prompt Engineering · Model Evaluation · Data-Driven
Luxembourg Centre for Contemporary and Digital History, University of Luxembourg
- Built and optimized a modular LLM-powered AI pipeline for automated entity and relationship extraction from unstructured text documents — integrating LangChain, OpenAI APIs (GPT-4o, GPT-4.1, GPT-5, o3), multi-stage preprocessing, and prompt engineering into a fully configurable, end-to-end NLP system.
- Engineered a dedicated coreference resolution module as an isolated, independently validated preprocessing layer — improving downstream extraction accuracy by ~20% through systematic prompt optimization and modular AI system design.
- Designed and executed quantitative benchmarking across 43 system configurations — varying chunking, context windows, prompting strategies, and model selection — evaluated using precision, recall, and F1-score against 176 labeled ground-truth relationships; achieved a 250% F1-score improvement through data-driven optimization.
- Validated system reliability and reproducibility via 10-run stability testing; applied statistical model comparison to identify OpenAI o3 as optimal for reasoning-intensive extraction — translating findings into actionable model selection and deployment guidelines.
- Research formed the basis of an interdisciplinary Master's thesis at the intersection of NLP and Digital History; findings are being prepared for publication in a peer-reviewed journal in collaboration with supervisors at C²DH.