Declarative GenAI-Enabled Logical Iterators for RDF Generation in RML [C. Debruyne]

This thesis proposes extending the RDF Mapping Language (RML) with a declarative, GenAI-enabled extraction step that produces an iterable logical source for RDF generation. The goal is to define and prototype “GenAI logical iterators” with clear semantics for how inputs (prompts plus supporting files or entity lists) are transformed into record streams that RML can map into RDF, while ensuring reproducibility, provenance, and constraint-based validation of the generated triples. The work will be evaluated through concrete use cases, such as systematically tagging movie clips with controlled-vocabulary annotations and materializing the results as RDF, assessing quality, scalability, and cost.