Abstract
This panel presents Nabi, an educational AI system developed by the Big Data Studies Lab in the Faculty of Arts, with the support of HKU’s Teaching Development Grant (TDG). Nabi integrates retrieval-augmented generation (RAG), chain-of-thought (CoT) reasoning, and knowledge bases co-created by students from the course readings and their annotations. Students also have the option to toggle the inclusion of teacher’s notes in the AI’s responses.
Nabi’s modular architecture draws knowledge from vector embeddings, producing context-aware responses using smaller, energy-efficient open-weight language model (LM). An alpha version has been piloted in PI Javier Cha’s two courses: HIST4037 (Automating the Past: Artificial Intelligence and the Historian’s Craft) and HIST2218 (Medieval and Early Modern Korea).
The centerpiece of Nabi is a navigable 3D “landscape of knowledge,” where related works cluster together. Our team has intentionally moved away from the minimalist interface seen in commercial generative AI tools like ChatGPT, Gemini, DeepSeek, and Grok, which encourage generic zero-shot prompts. Instead, Nabi invites students and instructors to work within a defined vector space relevant to their query, adjust output entropy (with a disclaimer that more “creative” responses may increase the risk of hallucination), and actively contribute to the system’s knowledge base with user-curated entries. The goal is to break down complex ideas into semantic patterns, clarify conceptual relationships, and offer a clearer sense of the intellectual terrain of the course materials—while demystifying the “black box” nature of generative AI and promoting more transparent, responsible uses of the technology.
In this team presentation, we will share how we designed, developed, and deployed Nabi, as well as the survey results and interviews conducted to measure its pedagogical impact.
Presenter(s) / Author(s)
