Enterprise Knowledge Search
Wikimedia Deutschland needed to make Wikidata’s massive, multilingual knowledge graph usable for modern generative AI applications because the existing access pattern depended on specialized query languages and complex retrieval workflows that limited natural-language exploration.
Reported impact
Wikimedia Deutschland created a faster, simpler way for AI systems and developers to explore open knowledge, making Wikidata more accessible for semantic search and future multilingual generative AI use cases.
Wikimedia Deutschland introduced a semantic retrieval architecture built on IBM DataStax Astra DB inside IBM watsonx.data so generative AI systems can search Wikidata with vector techniques instead of handcrafted queries. The design stores and retrieves embeddings across the large graph of entities and relationships that underpin Wikidata, allowing large language models to identify relevant facts and context quickly across languages and topics. By combining vector search with Wikidata’s existing structured graph, the platform gives developers and researchers a natural-language entry point into one of the world’s largest open knowledge repositories while preserving the precision that comes from the underlying linked data. The team is using this architecture as a base for expanding language coverage, scaling to larger datasets, and improving how AI systems discover, navigate, and interpret open knowledge, turning Wikidata into a more practical foundation for downstream generative AI tools rather than just a reference dataset for specialists.
Natural-language prompts are translated into semantic retrieval against embeddings stored in Astra DB within watsonx.data, relevant Wikidata entities and relationships are surfaced from the knowledge graph, and that context is returned to downstream generative AI applications and researchers for multilingual exploration.
“30x faster query performance”
“90% reduction in build time”
“Built on IBM DataStax Astra DB within IBM watsonx.data, the architecture enables large language models to quickly identify relevant entities and relationships across Wikidata’s massive knowledge graph.”
“Instead of writing complex queries, users can explore information through natural language prompts”
“The new approach allows developers and generative AI systems to retrieve relevant knowledge more easily”
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
IBM Sample Nov 6, 2024 ASPHI Onlus ASPHI Onlus needed a better way to make reading materials accessible for people with hearing or cognitive impairments and non-native Italian speakers because conventional text experiences created barriers to learning… 4 weeks IBM Cloud Code Engine · IBM Cloudant · IBM watsonx Assistant IBM Sample Jul 31, 2026 Crazylog Crazylog needed to help industrial maintenance technicians find trusted technical information quickly during equipment failures, even as documentation volumes, compliance needs, and workforce turnover made field… 10 frontline naval vessels IBM Cloud · IBM watsonx.ai · IBM watsonx.data IBM Sample Jun 24, 2026 trivago trivago needed a secure, EU-compliant way for employees to use generative AI across knowledge work without locking the company to one provider or exposing internal data to public tools, while also reducing the friction… 70% of the company attended the AI Learning Day IBM Digital Product Design and Engineering Consulting · trivago Copilot 