Vietnamese system automates privacy record extraction with local AI models
Automated Extraction of Records of Processing Activities (RoPA) Using Hybrid RAG and Locally Deployed Large Language Models
Computation and LanguageCryptography and SecurityInformation Retrieval
Summary
Many organizations in Vietnam must document how they handle personal data, but doing this by hand is slow and difficult. The authors created a system called RoPA Manager that automatically pulls out important information from documents using a mix of search techniques and AI language models that run on local computers. This approach respects privacy rules that restrict sending data to online services. They tested their system on real Vietnamese data and found it works well without errors, and the locally run AI models performed about as well as cloud-based ones.
What this means in practice
- •For privacy compliance teams: Automatically create accurate records of personal data processing activities without sending data to external cloud services.
- •For enterprise it teams: Deploy local AI models to extract regulatory data from documents to comply with data sovereignty laws efficiently.
Authors
To Duy Hinh, Nguyen Le Quoc Anh, Phan Van Tri, Khuong Nguyen-An
Abstract
Vietnam's Personal Data Protection Law (Law No. 91/2025/QH15) and Decree No. 356/2025/ND-CP, effective January 1, 2026, require organizations to establish and maintain Records of Processing Activities (RoPA). Manual RoPA preparation is labor-intensive, while cloud-hosted large language models (LLMs) may conflict with data-sovereignty requirements. We propose RoPA Manager, a system for automated RoPA information extraction using hybrid retrieval that combines lexical ranking over tsvector, dense-vector search, Reciprocal Rank Fusion (RRF), and locally deployed LLMs. We introduce a Vietnamese RoPA benchmark with 32 organizations, 77 processing activities, 12 field groups, and 4,338 reference values. Evaluation is reported at three distinct levels. The automated scorer, tested on perturbed data without invoking an LLM, achieved F1 = 0.9493 [0.9436, 0.9548]; this measures scorer robustness rather than end-to-end extraction accuracy. End-to-end extraction achieved token coverage of 50.04-55.25% against the reference labels. Two independent experts reviewed 1,558 reference values (35.9% of the benchmark), found no incorrect values, and achieved 99.68% agreement with PABAK = 0.9936. Value-level precision was not measured. Across 32 paired scenarios on a 24 GB GPU, locally deployed Qwen3.5-27B-GPTQ-Int4 showed no statistically significant difference from cloud-based DeepSeek-V4-Flash (difference 0.20 percentage points in favor of DeepSeek, 95% CI [-0.93, 1.32], p = 0.72), while Gemma-4-31B performed significantly worse (p < 0.01).