AI teaching · what it actually does

You describe how you want to teach.
The assistant follows it — and shows you what happened.

Uedu is an AI-mediated learning platform running in production across higher education in Taiwan. This page is the plain description of what you can run in your own course: no pricing table, no trial period, and no code to write.

29
higher-education institutions
4,300+
learners
420,000+
learning interactions
The short version

Three things, in order

Most AI teaching tools stop after the first box. The reason this platform exists is the third one.

You
Write the behaviour

In ordinary sentences, in your own language. Attach your own readings so the assistant answers from them and cites the page.

Your students
Work with it, inside the course

Not a separate chatbot on the side — it sits alongside the quizzes, worksheets, forums and recordings that make up the course.

You, afterwards
See the kind of thinking

Dialogue is classified by cognitive level across the semester. That is a different question from how often students logged in.

Step 1

Build the assistant

You describe the teaching behaviour you want in plain language. There is nothing to install and no code to write.

Course AI assistants

Each course gets its own assistant. You write its persona and guidance rules as free text — how it should answer, what it should refuse, which sources it should prefer. Students join with a course code.

Your own teaching materials

Upload PDFs, slides and documents to a course. The assistant retrieves from them and cites the file and page it used, so students can check the claim against your material rather than trusting the model.

Socratic and debate modes

Optional modes for a course. In Socratic mode the assistant leads with questions instead of answers; in debate mode it argues a position so students have to respond to it. Both work on topics you set for your own material.

Students build assistants too

Any student can create their own assistant with the same editor — writing the instructions, not just consuming them. Several courses use this as the assignment itself.

Step 2

Run the course around it

The assistant is not a standalone chatbot. It sits inside the coursework students already have to do.

Quizzes and worksheets

Generate drafts from your material, then edit them. You decide what ships — nothing is published to students without your review.

Discussion forums

Course and campus-wide boards, with an optional scoring scheme that weighs semantic quality and peer response rather than post count.

Surveys and consent

In-platform surveys with scheduled opening and closing. Research consent, where it applies, is versioned and withdrawable.

Class recording

Record audio, screen, or webcam during class; transcripts are generated afterwards. Sharing with students is off unless you turn it on.

An independent learning companion

Aida works across a student's whole programme rather than one course, and is deliberately Socratic: it asks rather than answers. Course assistants default to answering directly — the two are meant to feel different.

Step 3

See how students actually think

The part most tools skip. Usage counts tell you who showed up; these tell you what kind of thinking happened.

Bloom's taxonomy trends

Dialogue is classified by cognitive level and tracked across the semester, per student and per class. This is the measure behind several of our published results.

Learning portfolios

A per-student view across conversations, assessments and participation, exportable for the student's own use.

Cognitive test battery

Twenty-three research-grade cognitive tasks — attention, working memory, executive function — with millisecond timing, for courses and studies that need a measured baseline.

Learner profiling

Three open psychometric instruments (Holland RIASEC, IPIP Big Five, OEJTS) that students take for themselves. Results belong to the student, not to your gradebook.

Throughout

Things you keep control of

The decisions that usually get made for you by a vendor.

Which model runs your course

A free default is provided. A course can instead use its own API key for OpenAI, Anthropic or xAI models — your key, your account, your usage. Background analysis never runs on your key.

Blind model comparison

A teaching activity where students compare two anonymous models on the same question and vote before the identities are revealed. The resulting rating is student preference, not a capability benchmark, and is labelled that way to students.

10 interface languages

The platform interface is translated into 10 languages. Students converse in whatever language they write in.

Your data leaves by one door

Research export is de-identified on the way out, logged, and gated on an ethics approval. There is no second route and no informal copy.

Programmatic access

A public read API and an MCP server expose low-sensitivity data (public courses, institutions, publications) to your own tools. Student work is never exposed this way.

Not shipped yet

Uedu Brain

A self-built physiological sensing headset (EEG, fNIRS, PPG). Currently in design and fabrication — no data has been collected yet.

Per-channel knowledge bases for student-built assistants

Retrieval scoped to a single student-created assistant, isolated from courses and from other channels. Planned, not yet shipped.

Working together

Two ways in

The platform is free to use and is not a commercial product. Model usage is covered either by your institution's own API key or by donation-funded capacity.

Teach with it

Run a course on the platform. You get the assistant, the coursework tools and the analytics for your own class. No fee, and no obligation to take part in any study.

What we need from you: a course, a semester, and honest feedback.

Research with it

Analyse de-identified data through the export route, or design a new study on top of a running course. Co-authorship is the normal arrangement.

What we need from you: a question worth asking. For existing platform data you do not need to file your own ethics submission — see below.
The question everyone asks first

Do you need your own ethics approval?

Usually not — and this is the part that surprises people. The platform holds an umbrella research-ethics protocol, and data governance under it is the platform's responsibility rather than yours. Three cases, so you can tell which one you are in.

Analysing data the platform already holds

Covered. You do not file your own submission for this.

De-identified export of data already collected on the platform runs under the platform's umbrella protocol (NTU-REC 202507EM058). Consent, de-identification and the export route are the platform's responsibility, not yours. We provide a provenance statement you can cite in your methods section.

Collecting something new

Needs to be added to the protocol first. Plan lead time.

A new instrument, a new sensor, an external questionnaire, or any question put to students that the protocol does not already cover requires an amendment before it starts. Amendments are batched, so tell us early rather than at the point you need the data.

Your own institution and your target journal

Yours. We supply the paperwork you need to answer them.

Some institutions require their own registration or approval regardless of where data was collected, and most journals require an ethics statement. Those obligations sit with you — but they are usually satisfied by citing the umbrella protocol rather than by running a fresh review.

Research participation is always separate from platform use, and never affects a student's grade. Students who decline or withdraw keep full access to their courses. Full policy: research project governance.

Before you decide

What this is not

Worth reading before you invest a semester in it.

It is a research platform, not a product.
There is no sales team, no service-level agreement and no support contract. Issues reach a small team by email.
Most of the interface was written for Taiwanese higher education.
10 languages are available, but some documentation and teacher-facing guides are still Chinese-first.
Free is a funding arrangement, not a promise about the future.
Model usage costs real money. A course either uses its own API key or draws on donation-funded capacity, and that capacity is finite.
We publish what did not work as well.
The failure modes, the incidents and the measurement mistakes are in the papers and the tutorial materials, not hidden behind the demo.

Tell us what you teach.

One paragraph is enough: the course, roughly how many students, and what you want the AI to do. We will tell you honestly whether the platform fits — including when it does not.

[email protected]

Uedu is developed by Chia-Kai Chang, Assistant Professor at the Center for General Education, National Central University, Taiwan, and is adopted by institutions beyond it. Research on the platform runs under an umbrella ethics approval.

International overview · Publications · Developer API · uedu.tw