TAIHU is a dialogue-based knowledge exploration system tailored for Taiwanese humanities research. It integrates Taiwanese unique databases and applies Retrieval-Augmented Generation to retrieve precise information while minimizing AI hallucinations. The system features a fact-checking mechanism that highlights supporting evidence to enhance credibility and interpretability. A localized benchmark and user experience research further support system optimization and real-world applicability.

TAIHU system demonstrates four major breakthroughs: optimization of RAG technology, an explainable fact-checking module, a systematic user experience framework, and the development of localized benchmarks for Taiwanese humanities data. It ranked in the top 10% globally in the CRAG international competition. Our team is currently developing an agentic workflow, automated retrieval path planning, and semantic search methods to enhance the system’s response accuracy and overall performance.

TAIHU system can support applications in humanities research, education, cultural curation, and libraries. It offers natural language querying, fact-checking, and multi-turn dialogue to enhance research and learning. It can also serve as a smart guide and knowledge tool for cultural institutions, improving audience interaction and content connection. With modular scalability, it can be customized to institutional needs, accelerating the digital transformation of humanities services.