At INSA / TETRA Project, I built a Python pipeline that transforms French technical PDF reports into RDF knowledge graphs using layout-aware extraction, image–text retrieval, and vision-language captioning.
I benchmarked six cross-modal retrieval models on annotated image–text associations across five reports, then selected a model using macro-nDCG@5. I also evaluated document-image captioning configurations and built provenance-aware RDF graphs with TetraOnto and RDFLib.
At Celebal Technologies, I developed batch ETL pipelines in Azure Data Factory and Databricks, and built automated data-validation and data-quality checks. My research project reimplemented a GMatching-based framework for one-shot rare-relation link prediction using ComplEx embeddings.
