sosa-pii-ner-ae-v1.0.0
Regional PII named entity recognition model for UAE documents.
Part of the SOSA DevOps Privacy Filter — a local-first, privacy-preserving
AI runtime for developers. Weights are Apache 2.0. No cloud required.
Model summary
Fine-tuned from urchade/gliner_large-v2.1 on synthetic UAE PII data. Detects
Emirates IDs, Tax Registration Numbers (TRN), Tax Identification Numbers (TIN),
UAE mobile and workplace phone numbers in Arabic and Arabic-English mixed documents.
Intended use: Local PII detection within the SOSA DevOps Privacy Filter
sidecar. Text never leaves the user's machine.
Labels
Label Description Format Validator ae_emirates_idUAE Emirates ID (رقم الهوية) 784-YYYY-NNNNNNN-C (15 digits) Prefix 784 + birth year ae_trnTax Registration Number 15 digits Length validation ae_tinTax Identification Number 10 digits Length validation ae_phone_mobileUAE mobile phone 05X-XXX-XXXX (10 digits) Operator prefix (050/052/054/055/056/058) ae_phone_officeUAE office/fixed phone 0X-XXX-XXXX (9 digits) Area code validation
Global labels also carried (defence-in-depth):
email, phone_e164, credit_card, passport_generic, ipv4_public
Evaluation — v1.0.0 gate results
Label F1 Gate ae_emirates_id0.9425 ≥ 0.85 ✅ ae_trn0.8421 ≥ 0.80 ✅
First-run gate pass (T1). Training: D-AE-1 dataset, 10,000 steps, A40 GPU,
2026-05-28/29.
Limitations
TRN/TIN numeric collision: Both labels are long numeric sequences.
Context (VAT/TRN vs TIN/e-invoicing keywords) disambiguates.Arabic script context: Trained on Arabic and English keyword contexts.
Pure transliterated-only contexts may underperform.Context-gated: Bare values without surrounding context are unreliable.
Training data
Synthetic UAE PII examples only. No real resident or employee data used.
Arabic and English context templates included.
Integrity
pytorch_model.bin SHA-256:
c7ef875f4563519e3aab189aec244a2e444bb231e230404f68f6004ed5567be6License
Apache 2.0 — inherited from
urchade/gliner_large-v2.1.
Fine-tuned by Sovereign Systems.
See
LICENSE .
