Bridging the Metadata Gap: Automated VRE Provisioning via Agentic AI and CodeMeta Knowledge Graphs

Abstract

Virtual Research Environments (VREs) are essential for modern data-driven research, yet the manual overhead of configuring these environments remains a significant barrier for researchers in the Social Sciences and Humanities (SSH). This paper proposes a framework within the Macroscope project to automate VRE setup by leveraging Agentic AI for metadata generation. By populating a Knowledge Graph with CodeMeta descriptions generated automatically from software repositories, we enable a machine-readable pipeline that bridges the gap between raw data access and functional, containerized research environments.

Question we try to solve:

• Which tools and versions can manipulate my specific data?
• Are the software licenses compatible with my research stack and institutional requirements?
• What are the specific operating system, memory, and CPU requirements?
• Where is the documentation, and how should packages be sequenced during installation?
• Is there a citable publication or an ORCID associated with the authors?

How to run

Install Multipass to create your VMs locally (Experiments with Macbook):

• brew install --cask multipass

Install PyYAML: You'll need this for the Python script to read your configuration.

• pip install pyyaml

Execution

• python run_vm_rocrate.py

Ro-Crate experiments

run_vm_rocrate.py creates a ro-crate-metadata.yaml file and runs a VM configured by it.

You can use this ro-crate file to create a vm in other environments; there's no need to be multipass on a MacBook.

How it Works

The script leverages macOS's Hypervisor via Multipass. Here is a high-level look at how the layers interact:

• YAML Parsing: The script uses PyYAML to turn your human-readable settings into a Python dictionary.
• Resource Allocation: It maps the CPUs, memory, and disk keys directly to the command-line arguments that Multipass requires.
• Cloud-Init: If you want the VM to come pre-installed with software (like Git or Docker), the cloud_init section handles that automatically upon the first boot.
• Native Performance: Because it uses the Virtualization Framework (on Apple Silicon or Intel), there is very little overhead compared to heavier tools like VirtualBox.

Useful Management VM multipass Commands

Once your script has started the VM, you can manage it from your terminal. Here are some useful commands:

• Enter the VM multipass shell research-lab-env
• Check Status multipass list
• Stop the VM multipass stop research-lab-env
• Delete the VM multipass delete --purge research-lab-env