About Us
HallucinationNerd is a large language model-based tool that checks whether the sources cited in a piece of text actually support the claims attributed to them. Given any document with inline citations — a retrieval-augmented (RAG) answer, an agentic system's report, or a research paper's related-work section — it breaks the text into individual claims, retrieves the relevant passages from each cited source, and reports, for every claim, whether the source supports it, partially supports it, does not support it, or contradicts it. It works on PDFs, web pages, and plain text.
How It Works
HallucinationNerd runs a four-stage pipeline: it extracts atomic claims and their citations, retrieves the cited source's content and locates the most relevant passage, and then asks a structured LLM judge whether that source supports the claim, returning a per-claim verdict with an evidence quote. For claims that carry no citation, it can optionally search user-selected databases (PubMed, arXiv, Semantic Scholar) to find or refute supporting evidence. A configurable strictness setting lets users trade precision for recall.
Meet Our Team
Dennis Shasha
Professor of Computer Science
Courant Institute of Mathematical Sciences, New York University, New York, USA