Researcher in NLP and information retrieval, focused on making machines understand tables in scientific documents. Extraction, structure recognition, and entity linking in domain-specific literature. PhD from the University of Bamberg, research conducted at HITS Heidelberg.
My thesis spans three threads: digitising historical tabular records, IUPAC-aware chemical entity linking in STM corpora, and ML methods development for table extraction. The connecting thread is building systems that actually work on messy real-world documents.
BeeProject — OCR and computer vision pipeline for digitizing historical handwritten beekeeping records. The interesting part is handling severe layout variation and handwriting noise at scale. Published at JCDL'24.
FunChem — Annotated dataset of functional chemistry tables from STM literature (751 tables), with IUPAC entity annotation for table cell content. Coming soon.
FindChem — IUPAC-aware chemical entity linking system for scientific documents, built on HuggingFace Transformers. In progress.
Python · PyTorch · HuggingFace · Neo4j · Docker