An AI Twin of a Professor, Trained on Your Lectures
August 14, 2026 · 6 min read · by the hiy team
The short answer
Yes — you can build an AI twin of a professor trained on your lectures to answer student questions, and the setup takes an afternoon. Slides, papers and the syllabus go in as files. Recorded lectures come in from YouTube. Deadlines and grading policies go into a structured list so they are answered from exact rows instead of paraphrased. Every grounded answer carries a citation that opens the passage it came from, and when a question falls outside your material the twin says it doesn't know rather than guessing.
That last behaviour is the point of the whole exercise, and in a teaching context it is not a nicety.
Why a professor's twin has a higher bar than a creator's
Because a wrong answer lands on a student's record. If a marketing twin invents a testimonial, someone gets annoyed. If your twin invents an exam date, a late-submission window, or a citation for a paper that doesn't exist, a student acts on it — misses a deadline, hands in a fabricated reference, sits an exam on the wrong day. The harm is measurable and it falls on the person with the least power in the arrangement.
So the two features that read as polish elsewhere are load-bearing here. A designed "I don't know" means the twin has a floor it will not go below. A citation that opens the source passage means a student can check the answer against your actual slide before they act on it. Neither is optional in a classroom. Both should be non-removable, and in hiy they are: no plan takes away the citations, the honest refusal, the AI disclosure label, or the report link on a public twin.
If you sell courses to an open market rather than teach an enrolled cohort, the calculus is different and the course creator version of this post is the right one — that reader is optimising for conversion, and their failure mode is a lost sale, not a harmed student.
How do you train an AI twin on your lectures to answer student questions?
Six steps, in this order.
1. Files first: syllabus, slides, reading list, past problem sets. These are the highest-signal material you have and they are already written. Upload the syllabus first — it is the single most-asked document in any course.
2. Recorded lectures via YouTube. Point the twin at your channel or at individual videos and it reads the transcripts. This is where the conceptual explanations live: the way you explained eigenvectors, not the way a textbook does.
3. Deadlines and policies as a list, not prose. This is the mechanism that keeps the twin from inventing a date. A list is a small structured table with fields you define — assessment name, due date, weighting, late penalty, extension route — and the twin answers from exact rows rather than from a paraphrase of a paragraph. If your department publishes dates as a CSV or JSON feed, a list can refill from that feed on a schedule so a moved deadline propagates without you re-uploading anything.
4. Your Cal.com or Calendly link, so "see a human" is one click. hiy has no calendar of its own — it uses the booking link you already have. The value is the handoff: a question about a grade dispute, an accommodation, or a personal circumstance should become a booked office hour, not an answer.
5. Test chat before anyone sees it. Paste in the twenty questions you actually get every term — the ones already in your inbox. Watch what it does with the three you'd hate it to get wrong. This sandbox costs nothing and counts against nothing.
6. Publish, then put the link where students already are. Publishing hands you a launch kit: the link, a QR code, and three ready-to-send posts. The QR code belongs on your first lecture slide and in the syllabus; the link belongs on the LMS page, not in an announcement that scrolls away. If your LMS page accepts HTML, you can embed the twin directly in it — inline, or as a bubble that loads nothing until a student opens it.
Which questions should go to the twin, and which shouldn't
Anything answerable from an exact row or a passage you wrote goes to the twin; anything that needs judgment, discretion or a duty of care comes to you.
| Question type | Where the answer lives | What the student gets |
|---|---|---|
| "When is the essay due? What's the late penalty?" | List row | The exact row, unparaphrased |
| "Can you explain the thing you said about selection bias?" | Lecture transcript, slides | An answer in your framing, with a citation that opens the passage |
| "Which chapter covers this?" | Reading list, syllabus | A pointer, cited |
| "Is X on the exam?" | Only if you wrote it down | An honest "I don't know" if you didn't |
| "Can I have an extension? I've been ill." | Nowhere, and correctly so | Your booking link, and a booked office hour |
The last two rows matter more than the first three. A twin that answers everything is worse than one that answers most things and hands the rest to you with the student's context intact.
What happens to the questions it can't answer
They queue for you. When the twin hits a gap it tells the student it doesn't know, shows the searches it ran first — their question, then its own rewordings of it — and files the question for your review. You answer once, and that answer is taught permanently; corrections you write override older sources, so if a policy changed mid-term your correction wins over the slide that contradicts it.
Over a term this is the actual mechanism of relief. Early on you'll get a long queue, because the twin only knows what you uploaded. Later the questions arriving in it are mostly ones worth your time.
The honest version
Some limits you should know before you set this up.
It is text only. hiy has no voice and no video avatar. It reads your lecture recordings and answers in writing. If what you want is students hearing your voice explain something at midnight, this is not that product. Voice is listed as something we're exploring for later, not something you can switch on now.
"Private" means unlisted, not gated. A private twin is kept out of search, sitemaps and public listings — but it is a link, and a link can be forwarded. Access restricted to an enrolled cohort is not something hiy does today. Assume anything you put in is readable by anyone who receives the URL, and keep student data, grades and anything identifying out of the sources entirely.
It does not re-read your sources on its own. Upload new slides when you make them. Lists are the exception, since they can refill from a feed.
Turn off lead capture. New twins ask visitors for contact details after a couple of messages, which is right for a consultant and wrong for a lecturer whose students are already enrolled. One tap in settings switches it off. Do it before you publish.
You may only twin yourself, or a product or organisation you own. There is deliberately no "with their permission" carve-out, so you cannot build one of a colleague, a co-lecturer, or a departed emeritus, even with their blessing. It's your material and your name or it doesn't get made.
Where to start on Monday
Upload one document: the syllabus. Then open the test chat and ask it the five questions in your inbox right now. You'll know within ten minutes whether the rest of the term's material is worth adding — and if you teach more than one course, each one gets its own twin with its own sources and its own rules.
The free plan includes a real published twin, and students always chat free. Build one free, or read how sources work first if you'd rather see the shape of it before uploading anything.