TAIHU is a dialogue-based knowledge exploration system designed for humanities research in Taiwan. It integrates multiple unique full-text primary databases. By adopting Retrieval-Augmented Generation (RAG), the system leverages designated sources to reduce the risk of AI hallucinations. It also incorporates fact-checking and source attribution to enhance accuracy and transparency. In addition, the team has developed an AI benchmarking framework and introduced user research processes to advance the system. These efforts give TAIHU strong potential for both academic and societal applications.

This system integrates unique Taiwanese humanities databases. It overcomes the limitations of web search and general large language models by creating a local knowledge network through deep semantic understanding. Its core innovation is a "sentence-by-sentence generative source verification" mechanism, which resolves knowledge gaps in general models on local topics and reduces the risk of AI hallucinations. Furthermore, the team has established the first AI benchmark for Taiwanese humanities to objectively evaluate system performance. This achievement contributes to building a reliable and efficient digital knowledge platform for humanities research.

TAIHU can be widely applied in humanities research, educational settings, and a variety of cultural and performing arts institutions. It supports natural language queries, multi-turn dialogues, and source verification functions. These features enhance research efficiency and the learning experience. It can also serve as an intelligent guide and knowledge-sharing tool, strengthening audience engagement. The system is designed with modular scalability and can be customized to meet institutional needs, supporting the digital transformation of humanities knowledge services.