AI-Enabled Curation Assistant (The Sax Institute)

Overview & Background

Intersect’s 3AI collaborated with the Sax Institute to develop a proof-of-concept AI-Enabled Curation Assistant designed to be able to be implemented within the Secure Unified Research Environment (SURE*). The project aimed to automate operational bottlenecks and reduce the manual burden on data custodians reviewing research outputs for download from SURE, while managing high volumes of data ingestion.

The system provides AI-assisted diagnostic tools for risk assessment, effectively identifying statistical disclosure risks, such as Personally Identifying Information (PII) and small cell sizes, to enhance data security and operational efficiency.

As a proof-of-concept project, we expect further development and testing of the proof-of-concept AI-Enabled Curation Assistant to ensure its security, utility and acceptability with data custodians and researchers before deployment to the live SURE environment.

Technical Execution & Solution

The solution combined LLM reasoning with traditional machine learning and automated data pipelines, all while maintaining the SURE security model and protecting against any data egress outside the SURE perimeter:

  • LLM-based Risk Assessment: Utilised advanced large language models for automated reasoning and disclosure risk assessment.
  • Traditional machine learning and rule-based programming: Integrated automated PII detection tools, traditional machine learning models and rule-based programming output into the LLM for more predictable and consistent outputs.
  • Data Orchestration & Database Design: Orchestrated workflows using automated orchestration tooling with a relational and vector database design deployed via containerization and cluster management infrastructure.

Impact & Research Results

This solution reduces operational bottlenecks in the curation process, significantly increasing operational efficiency and timeliness. Further, the proof-of-concept shows the ability for enhanced data security through augmented risk assessment that identifies statistical disclosure risks, including PII and small cell sizes, while delivering a scalable research infrastructure with modular design.

This proof-of-concept has paved the way for development of an optional tool for data custodians and researchers to elect to use within the SURE Curated Gateway.

*SURE is operated by the Sax Institute and is supported by Data Connect Australia which is an initiative of the Australian Government National Collaborative Research Infrastructure Strategy (NCRIS) and the NSW Government.