PoliSage is an intelligent policy question-answering system powered by large language models (LLMs) and the Retrieval-Augmented Generation (RAG) framework.
PoliSage is an intelligent policy question-answering system powered by large language models (LLMs) and the Retrieval-Augmented Generation (RAG) framework.
PoliSage is an intelligent policy question-answering system powered by large language models (LLMs) and the Retrieval-Augmented Generation (RAG) framework. Unlike traditional LLMs that may generate hallucinated or outdated content, Polisage grounds all responses in authoritative policy documents retrieved in real time. It uses semantic search to locate relevant text and then leverages LLMs to generate clear, human-like answers with source citations. This hybrid approach combines the fluency of LLMs with the factual accuracy of retrieval, ensuring trustworthy, context-aware outputs. Supporting multilingual input and document traceability, Polisage maximizes the effectiveness of LLMs in high-stakes environments where policy clarity and reliability are critical.
Polisage provides responses grounded in official policy documents, ensuring every answer is verifiable and eliminating hallucinations common in traditional language models.
By combining semantic retrieval, keyword matching, and large language model generation, Polisage delivers accurate, relevant, and fluent answers, even for complex or cross-document queries.
Polisage supports both Chinese and English inputs, adapts to diverse policy domains such as academic rules, housing guidelines, and administrative procedures, and its modular design enables fast updates and scalability.
The user query is first encoded into a vector and processed through both vector search and keyword matching to retrieve the most relevant policy document pieces. These retrieved fragments, together with the question, are then fed into a large language model to generate a coherent response. Finally, the system delivers an answer with clearly annotated sources and provides links to the original policy documents for verification.
Email nancyli@ln.edu.hk for more information about this project