How we handle lesson records

Last updated: 30 August 2026

UPCLASS keeps a record of what happens in a lesson so that we can make the next one better. This page describes what we keep, what we use it for, and what we do not do — as the system actually works.

1. What we keep

The material and page opened in class, words looked up, sentences translated, chat messages, and when the whiteboard was used.
Speech turned into text by speech recognition. This is temporary working material, not a lasting record (see section 4).
Microphone audio during the lesson (from 30 August 2026). Each device records only its own microphone, in 30-second clips, and encrypts them before sending. A device never records the other person. The audio is used only to produce text, and is deleted within 24 hours (see section 4).
The teacher's four scores (speaking, grammar, vocabulary, attitude) and the teacher's notes.

2. What we use it for

To generate the lesson report automatically (words, grammar, corrections, practice questions).
To draft the next lesson plan. Anything the teacher edits is never overwritten by the AI.
To count where a learner repeatedly stumbles, shown on the teacher's screen only — not the learner's. A running total of mistakes is a report card, not feedback.

3. Where the processing happens

Reports are generated by a language model running on the school computer. Lesson conversations are not sent to any external AI service.
Speech is also turned into text on the school computer (Whisper). No external transcription service is used.
Audio is held in cloud storage only while it is on its way to the school computer, and only encrypted — the key exists solely on the school computer, so the cloud provider cannot open it.
Reports and records are stored on the school computer. Booking and billing data is held separately in a cloud database.

4. What we delete

The audio itself is deleted within 24 hours of being transcribed. Only clips that failed to transcribe are kept, so they can be retried — once deleted they cannot be recovered.
Speech transcribed by speech recognition is deleted once the report is more than seven days old and has been confirmed to have substance.
This is so that conversations do not remain indefinitely. It is not written into the local archive either.

5. We may use it to write about learning (added August 2026)

We sometimes write about patterns — such as where Japanese learners tend to stumble — on our school blog.
When we do, we use only aggregated figures that cannot identify an individual. We do not use names, IDs, contact details, teacher names, teacher scores, or the sentences a learner actually wrote.
We also publish only what appears across several different learners. Anything specific to a small number of people is excluded automatically, because anonymising it is not enough to keep that person unrecognisable.
Where an article gives an example of a learner, it is a composite drawn from several people, and the article says so. We do not retell any one real person.
If you would prefer your records not to be used this way, please tell us and we will exclude them.

6. What we do not do

We do not sell or pass records to third parties.
We do not show the teacher's private notes or scores on a learner's screen or report.
We do not record lessons on video.

7. Contact

For any question or request about these records, please tell your teacher or contact us on LINE.