Hallucination-free Intelligent Research Assistant with Knowledge Graphs

Verbatim is an AI research platform for hallucination-free question-answering based on trusted documents. Our open-source technology grounds its answers directly in source texts, but it does not currently interface with structured data sources such as knowledge graphs (KGs). This project will investigate how the Verbatim approach can be extended to extract relational triplets from natural language, a key step in mapping information to KGs. In collaboration with the WU’s SemSys group we shall also explore the use of this novel technology in the construction of high-quality KGs for research, as well as the potential of existing KGs in validating knowledge extracted from text documents. A research prototype for knowledge-based question answering shall be developed and extensively tested with researchers across multiple disciplines, laying the foundation for a major new feature of Verbatim that shall enable maximum verifiability and auditability of AI-assisted academic research.