KVKK Masker
Personal and sensitive personal data, found and masked in Turkish text — with the KVKK article behind each detection named.
In development
What it does
- Rules plus a small model
- A rule and pattern layer runs alongside a compact language model. The rules catch what has a shape; the model catches what needs reading.
- Names the article
- Each detection is reported with the KVKK article it falls under, so the output can be reviewed against the law rather than against a generic entity label.
- Masked in place, reversibly
- Masks are applied at exact character offsets and the mapping is kept, so text can be unmasked by someone holding the key rather than destroyed.
- Small enough to run anywhere
- The model is deliberately small, so the whole thing runs on modest hardware — including inside a perimeter that no API call leaves.
KVKK Masker
- Language
- Turkish
- Detects
- Personal and sensitive personal data
- Output
- Masked text, spans, KVKK article per detection
- Status
- In development
Used for situations like these.
Not an exhaustive list — the shape of the problem matters more than the industry it sits in.
- Sanitising support transcripts, tickets and chat logs before they reach an analytics tool
- Preparing Turkish document sets for model training or third-party processing
- Redacting court records, contracts and official correspondence
- Screening system and application logs that may carry personal data
A detection, not a legal assessment
The output identifies what appears to be personal data under KVKK. It is not legal advice and does not constitute a compliance determination. Treat it as a control that reduces manual review, reviewed by someone who owns that judgement.