QueryPie releases QEV model for sensitive data classification
QueryPie says QEV detected 99.69% of sensitive documents in its own evaluation, missing 4 of 1,281 versus 46 for JEV API.
QueryPie announced on November 2, 2026 that it has developed QEV, a decision-making AI model that classifies personal information and internal confidential data, and published it on Hugging Face. The company said a demo supporting Korean, English and Japanese input is also available on the platform.
In QueryPie's own evaluation using Korean, English and Japanese documents, QEV detected 1,277 of 1,281 sensitive documents, a detection rate of 99.69%, according to dailysecu.com. On the same set, the commercial decision-making model JEV API detected 1,235 of 1,281. QEV missed 4 documents and JEV API missed 46, which QueryPie put at about 91.3% fewer missed detections for its model.
QEV scored 98.98% on the F2 metric, which weights recall more heavily than precision. The model sorts input into four categories: no sensitive information, personal information included, internal confidential information included, and both included.
In a demonstration using 10 fictional emails and AI prompts, QEV classified customer names, contact details and shipping addresses as personal information, and pre-release pricing and discount terms as internal confidential information.
QueryPie said it applied INT8 quantization to reduce GPU memory usage of QEV models by 27.5% to 45.2%. After quantization the model retained its 4 missed detections and its F2 score moved from 98.98% to 99.02%, the company said.
Quick answers
What is QEV?
QEV is a decision-making AI model developed by QueryPie that classifies personal information and internal confidential data.
Where can I try QEV?
QueryPie published QEV on Hugging Face along with a demo that supports Korean, English and Japanese input.
How accurate is QEV?
In QueryPie's own evaluation of 1,281 documents, QEV detected 1,277, a rate of 99.69%, and scored 98.98% on the F2 metric.