We have upgraded MatPortal's backend with a dense vector retrieval engine and GraphRAG, allowing the assistant to traverse ontology hierarchies for highly accurate domain context.
We've deployed an entirely new retrieval pipeline built on Weaviate and LangChain. When you query the AI Assistant, it doesn't just search text—it actively explores the semantic graph of the active ontology. This means our Deep Agents can now understand subclass relationships, property restrictions, and cross-domain mappings when constructing proposals.
Traditional search often misses context. By integrating the graph hierarchy into the prompt context, we've reduced hallucination rates and improved the accuracy of the assistant's suggested mappings and SPARQL templates.