Use of Roasting Data for AI Training

Operator. 커피익스플로러 provides 로스트파이. Inquiries may be submitted through Customer Support or by calling 070-8628-9532.

Purpose. Training and validation of a shared AI model offered to multiple customers, and development of Roast Pi analytics features. Development results may also be used in paid features.

Data fields. Recorded fields among temperature and RoR curves; elapsed times of roasting events; charge and discharge weights, weight-loss rate, and color; green-bean moisture and density; and a random analytics code. Temperatures are bean temperature (BT) and exhaust temperature (ET). RoR includes the rate of rise for those temperatures and any RoR included in the record. Event times are elapsed times to yellow and first crack. Heat-input and airflow time series and equipment manufacturer and model are excluded from the training fields.

Covered records. Records first saved to the cloud after consent. This includes older records that are first uploaded after consent. It excludes records already stored in the cloud before consent.

Safeguards. Names, email addresses, business names, customers, costs, blend ratios, and free-form notes are not used for training. Original profiles are not disclosed or sold to other customers. Authorized development personnel may review pseudonymized records. Training records are managed as pseudonymized and are not represented as fully anonymous. The mapping information connecting the random analytics code to the account is stored separately from training fields. Re-identification is not attempted.

Retention. Up to two years from the first training-data collection for each record. Use stops earlier, and training copies are deleted without delay, when the purpose is fulfilled, consent is withdrawn, or membership ends. Extracting the same record again does not restart the first-collection date. Two years is the retention period for training copies; it does not mean that a completed model is automatically destroyed after two years.

Permission to use. Users retain their data-related rights. The Company uses the data non-exclusively for the stated purposes, period, and scope. This does not permit an outside provider to train its own general-purpose AI.

Optional choice. You may sign up and pay without consenting, with the same plan features and prices. You may withdraw at any time under Settings → Data Use. The consent time, withdrawal time, and document version are recorded on the account.

Current collection. Publishing this proposal does not enable collection. This document does not state that collection is operating in production.

Withdrawal and renewed consent. Withdrawal in Settings stops new extraction and later training input. If consent is given again, coverage resumes with records first saved to the cloud after that time, and deleted training copies are not restored. A wording-version change alone does not erase the original consent time.

Deletion scope. Deletion and stopping apply to the contents of training copies and queued jobs intended to create them. The current collection path does not separately accumulate validation data or preprocessing derivatives. Original roasting records that a User stores in the Service are distinct from training copies and are not deleted merely because consent is withdrawn.

Coverage determination. If the same client record identifier existed on the server before consent, deleting and re-uploading that record does not make it eligible for training. A record is excluded if it has no identifier or the system cannot determine whether it was stored before consent. The time at which a new row is created is not enough by itself to make a record eligible.

Backups. The current retention setting for automatic database backups applies only to automatic snapshots. That period is not extended to other kinds of copies. While a training copy remains in a backup, withdrawal alone is not treated as having removed that copy as well.

Completed models. The Company does not guarantee that the influence of a specific record disappears immediately and completely after a shared model is created. If User information is reproduced in model output, the Company will investigate and consider removal, retraining, or discontinuing use. Two years is the retention period for training copies, not an automatic destruction deadline for a completed model.

General processing of personal information is governed by the Privacy Policy. Optional consent may be withdrawn under Data Use in Settings.