Uedu: A Mindfully Designed AI-Assisted Learning Platform Across Four Universities in Taiwan

Abstract

Generative AI has spawned a new generation of AI tutoring systems, but most deployments treat the AI as an isolated conversational agent, decoupled from the learner's state, the classroom context, and institutional constraints. This fragmentation limits the potential of technology-enhanced learning (TEL) to support the whole learner. We demonstrate Uedu, an AI-assisted learning platform built on a principle of mindful integration: every sensing modality is paired with a consent mechanism, every AI adaptation is transparent to the user, and every data flow is placed under ethical governance — echoing the EC-TEL 2026 theme, "Mindful TEL." The demonstration walks through six subsystems that map onto the six dimensions of Educational Omics, shown in real use across four universities in Taiwan.

Problem & Motivation

Most generative-AI tutors are deployed as standalone chat agents, disconnected from what the learner is doing, the course they are in, and the institution that hosts them. That disconnection caps what TEL can do for the whole learner — a system that answers a typed question but ignores the student's cognitive, physiological, social, and environmental context. Uedu's design response is mindful integration: each modality comes with consent, each AI adaptation is transparent, and each data flow is governed. Its theory of change is that mindful integration can deliver personalization without becoming surveillance.

Method

Uedu is built as a modular monolith in Python Flask from sixty-plus blueprint modules, with a companion native iOS app. Six subsystems map onto the six dimensions of Educational Omics: Uedu Core — the ClassroomGPT course assistant with a RAG knowledge base, real-time Bloom-level classification, knowledge graphs, AI quiz generation and Socratic-dialogue assessment, and EMI bilingual subtitles; Lecture Capture — synchronized screen/audio/webcam recording feeding Whisper transcription, dual-host podcasts, and personalized worksheets; Uedu Fit & Mind — Garmin wearable physiology and the PALM physiologically-aware language model, with an AI privacy panel; Uedu Sense — IoT environmental sensing (light, temperature, humidity, CO₂, noise); Learner Profiling — Holland RIASEC, IPIP Big Five, and OEJTS as voluntary self-exploration instruments; and Uedu Lab — a governed, ethics-approved data-export interface positioned as a trusted educational data lake.

Findings

• From lecture to personalized review: an instructor records a class (screen, audio, and webcam) and the system automatically produces a transcript, a dual-host podcast, and personalized worksheets, adding bilingual subtitles for EMI courses.
• Context-aware AI tutoring: a student asks ClassroomGPT a question; the system retrieves course content via RAG, references the learner's Bloom-level trajectory, and — with consent — incorporates PALM physiological context, while the teacher dashboard shows the Bloom distribution and knowledge-graph coverage.
• Learner agency: in the AI privacy panel, students preview "what will be shared" in natural language rather than raw values before toggling each feature individually; the same controls are available in the native iOS app.

Implications

Under continuous development since 2023, Uedu has accumulated tens-of-millions-scale AI message interactions through intensive use at four Taiwanese universities, demonstrating long-term, large-scale use in authentic settings. Several observations stand out: institutional onboarding requires only a subdomain configuration, with 50+ institutions registered with zero code changes; the RAG knowledge base and AI quiz generation are the most-adopted teacher features, directly reducing preparation burden; and students do actively use the AI privacy panel, indicating that learner-agency mechanisms are valued in practice rather than merely in principle.

Citation

@inproceedings{chang2026uedu_mindful_platform,
  author    = {Chia-Kai Chang and Kuei-Hao Li},
  title     = {Uedu: A Mindfully Designed AI-Assisted Learning Platform Across Four Universities in Taiwan},
  booktitle = {Proceedings of the 21st European Conference on Technology Enhanced Learning (EC-TEL 2026), Demonstration Track},
  address   = {Valencia, Spain},
  publisher = {Springer},
  series    = {Lecture Notes in Computer Science},
  year      = {2026},
  note      = {in press},
}