The course platform where the AI has actually read your materials.
A teacher owns a room, uploads their course materials and records the lecture. Students get the transcript, a summary, the assignments that were mentioned out loud — and an AI that answers their questions from that course’s real materials, not the open internet.
“Otter transcribes. Kru grounds every answer in the teacher’s own course materials, per course, across a whole semester.”
You can switch perspective at any time with the Teacher / Student toggle in the top bar.
Teacher records and uploads
The teacher owns the room. They record the lecture — voice, plus their screen if they present from a laptop — and upload the course materials.
AI turns it into notes
Shortly after class the transcript, a summary, key points and any assignments the teacher mentioned out loud are ready.
Students ask, grounded
Questions are answered from that course's own materials and transcript — with a citation to the page or timestamp it came from.
Built per faculty, because the vocabulary is the hard part
Each faculty speaks a different language. A general transcription model hears “lisinopril” as “list of April”. Grounding per course is what makes the answers usable.
lisinopril · bradykinin · angioedema · cardioselective
amortised · hash collision · memoisation · PDPA
WACC · beta · market risk premium · hurdle rate
What this prototype is — and is not
Everything here is clickable and fake, on purpose.
- · No microphone and no screen are accessed. Recording is a timer.
- · Uploaded files are listed, never read or processed.
- · No model runs. The AI answers were written by hand.
- · Nothing is saved. A refresh resets the demo.
Roadmap — not built yet
- Near-live view during class — Transcript and a rolling summary appear a few seconds behind the teacher, so students can follow along on their phone and catch what they missed in real time.
- Real transcription and real grounding — Whisper for transcription, and retrieval over the teacher's actual uploaded materials so every answer is generated from — and cites — the real course documents.
- The engine behind the Glossary tab — Each room's Glossary tab shows what this would produce — terms mined from the teacher's own documents, correcting "list of April" back to "lisinopril". Building the extraction and the correction pass, and measuring it against a hand-corrected reference transcript, is the project's main technical contribution and evaluation target.