You will be tasked with building a hallucination detection demo for retrieval-augmented generation (RAG) systems. Large language models sometimes produce confident-sounding but factually incorrect statements—a phenomenon known as hallucination. In RAG systems, an LLM generates answers based on retrieved source documents, but may still produce claims not grounded in any source. Your task is to build a tool that compares each claim in an AI-generated response to the retrieved documents and classifies it as supported, partially supported, or unsupported. You will work with existing open-source tools (such as LangChain) to set up a basic RAG pipeline, then implement sentence-level attribution using text similarity techniques such as embedding-based semantic matching. You will apply the methodology to a curated document collection (e.g., Wikipedia articles on a specific topic) and evaluate detection accuracy on sample question-answer pairs. Furthermore, you will create an interactive demo with color-coded trust indicators, conduct an in-depth analysis of false positives and false negatives, and propose potential improvements to the detection pipeline such as confidence thresholding or multi-source verification.