Research AI Sandbox

The Sandbox is a governed research platform designed to give computational researchers and health data scientists secure, privacy-protected access to leading-edge generative AI capabilities. Purpose-built for research workflows, the platform provides:

• Frontier Multi-Model Access: Direct access to advanced reasoning engines – including models from Anthropic, OpenAI, and Google – through a single secure gateway.
• Isolated Analytical Environment: A secure execution container that enables researchers to run Python and R code directly against local research datasets without exposing sensitive information.
• Multi-Document Synthesis & Literature Grounding: Advanced semantic search and synthesis across large academic literature collections, clinical evidence corpora, and complex datasets.

Current Phase: Closed Calibration Pilot

The Sandbox has launched as an initial closed calibration pilot with a selected cohort of ~185 computational power users spanning university departments and partner medical research institutes, including WEHI, Peter MacCallum Cancer Centre, Murdoch Children’s Research Institute (MCRI), The Florey, The Doherty Institute, St Vincent’s Institute (SVI), and The Royal Melbourne Hospital (RMH).

During this pilot phase, real-world research workflows are being evaluated to optimize platform performance, establish operational metrics, and refine user guidelines for advanced research computing across the precinct.

Looking Ahead: Platform Expansion

Following the conclusion of the calibration phase, usage patterns and technical telemetry gathered during the pilot will inform plans for broader platform access. As we learn more about demand, token consumption, and specific health data workflows, the platform will be progressively opened to a wider community of researchers, data scientists, and clinicians across the Melbourne Biomedical Precinct.

Responsible AI & Data Governance

Patient privacy and data security are central to the platform’s design. The initial pilot operates strictly under approved Public (“Green”) and Internal/Restricted (“Yellow”) research data classifications. Un-consented clinical patient data and highly sensitive medical records remain strictly prohibited while baseline governance protocols are established.

Further information

For further information about the Research AI Sandbox or broader initiatives across the Melbourne Biomedical AI Consortium, please visit the Research AI Sandbox SharePoint Hub (University of Melbourne login required) or contact Professor Andrew Lonie ([email protected]).