Enterprise AI must be not only correct but secure, fair and auditable. Ilmek products are developed and operated under the framework below.
Data and privacy
- Customer data is not used to train general models; it is used only for customer-specific fine-tuning, with explicit consent.
- Personal data can be masked at input (name, national ID, phone, IBAN).
- Data is stored in Türkiye, encrypted with AES-256; retention is set by the customer.
Reliability
- Generative answers come only from the organisation’s sources and cite them; if it is not in the sources, the system says ‘I don’t know’.
- Below the confidence threshold, the case hands over to a human.
- Regular red-team tests are run for prompt injection, data leakage and unauthorised access.
Fairness and audit
Group-level performance gaps are monitored in decision-support models. All requests and responses are written to the audit log; model versions and evaluation results are documented in the model card.