Philippe Saadé
Philippe Saadé is the AI/ML Project Manager at Wikimedia Deutschland, where he's developing a Vector Database on Wikidata's data to enable semantic search and support the open-source AI community in building projects with Wikidata's open data. He studied computer science at the Lebanese American University in Lebanon and completed a Master’s in Informatics at the Technical University of Munich in Germany, specializing in Machine Learning and Natural Language Processing.
Session
This presentation introduces the Model Context Protocol (MCP), an open-source standard for integrating AI models with external tools and data sources. We present a Wikidata MCP server that provides LLMs with core functionalities including semantic and keyword search for entity discovery, property exploration, relationship retrieval, and SPARQL query execution. This approach addresses key AI limitations by minimizing identifier hallucinations and incorrect assumptions about Wikidata's structure in tasks such as SPARQL query generation.
We also present a Wikidata vector database that enables semantic search across Wikidata's data, allowing LLMs to discover conceptually similar items even when exact terminology is unknown.