Uedu is an open educational research platform built at National Central University (Taiwan). We bring AI tutors, learning analytics, and physiological signals together — and we're at SDS 2026 looking for collaborators in the international data science community.
Wrist-worn HRV is enticing for educational researchers — non-invasive, continuous, deployable. But what does it actually measure once a real lecture, a real programming task, and six real undergraduates are in the room? We report a feasibility study with honest limits and one methodological warning we think every learning-analytics team should hear.
We assess whether Garmin vívoactive 5 watches — paired with the real-time Garmin Health Companion SDK (not the summary-only Garmin Health API) — yield consistent HRV during authentic Python programming classes (six undergraduates, two 3-hour sessions). Per-student ICC(3,1) between sign-aligned, z-scored HRV and Garmin's stress index was moderate-to-good (RMSSD .48–.74; pNN50 .64–.85; lnHF .54–.70). Group-level pooling of raw HRV produced strong but spurious event effects that vanished after person-centering — a Simpson-paradoxical pattern that should change how this community pools data. As a hypothesis-generating finding, smaller LF/HF increases during programming tracked larger midterm gains (ρ = −1.0, n = 4, descriptive). The honest message: a wearable is a measurement instrument, not a stress oracle, and person-centered analysis is a methodological prerequisite, not a stylistic choice.
Uedu is not a product. It is the live infrastructure for a research lab — running in production across multiple universities, instrumented end-to-end, and designed so that every interaction becomes a data point that students, teachers, and researchers can trust.
AI teaching assistants, classroom GPTs, quizzes, surveys, course management — running in live university classrooms across multiple campuses.
Garmin Connect integration with HRV, sleep, stress, and synchronized multi-person beat-to-beat interval streaming.
PALM — physiologically-aware language models that adapt tutoring to the learner's measured state.
IoT environmental sensing: light, temperature, humidity, noise, CO₂ — the room as a co-author of learning.
In-house wearable combining EEG, fNIRS, and PPG. Built for translational neuro-educational research.
The trusted educational data lake — researcher-facing exports, IRB-aligned access, reproducible cohorts.
We borrow the ‘omics’ vocabulary from the life sciences to argue that learning, like a living system, is only legible when you can read multiple modalities at once. Educational Omics is the framework that organises Uedu's data; it is also our case for why ed-tech needs a layered, hypothesis-aware research substrate, not just dashboards.
Cognitive trajectories — LLM dialogue, Bloom's taxonomy, reasoning traces.
Language complexity, semantic shift, prosody, automatic speech recognition.
HRV, sleep, stress, EEG, fNIRS — the body's running commentary on learning.
Forum interaction, collaborative writing, peer review, community formation.
Light, temperature, noise, CO₂ — the room as a quiet co-author of attention.
Consent, privacy, AI bias auditing, equitable access — first-class data, not a footnote.
Uedu is built and operated by an academic research group; every layer is documented, IRB-aligned, and designed to be cited in a paper rather than to win a pitch deck.
Recent venues where Uedu research has appeared or will appear:
Whether you're an academic, a data engineer, or a startup pushing AI into education — Uedu is open to thoughtful collaboration. We especially welcome partners across Asia–Pacific, Europe, and the international LAK / EDM / AIED communities.
Multimodal classroom data, federated fairness across institutions, physiological learning analytics, cross-cultural validation. We're actively recruiting partners for Apple Health × Garmin federated studies and for our group-level BBI synchrony work.
Email Chia-Kai
Uedu exposes a clean RESTful API at uedu.tw/api/v1 and a Model Context Protocol
server for LLM-native tooling. Free for academic research. Six tools cover universities, public
courses, course statistics, papers, and conferences — built so MCP-enabled agents can reason
over our published metadata.
Synchronized beat-to-beat interval streaming across an entire classroom, raw sequences preserved, group-level aggregation views available. Looking for collaborators in affective computing, physiological synchrony, and group flow research — and for partners who'd like to instrument their own cohorts.
Discuss with usIf any of this resonates, please find us at SDS 2026 — or reach out before the conference closes.