Hi Angelo,

We’re currently in the middle of building out an “Air-gapped”  PoC that integrates domain-specific AI agents into our Nutanix infrastructure workflows, specifically the “Nutanix Design Assistant” and I wanted to share what we’re aiming to achieve and see if anyone’s already gone down a similar path.

The goal is to improve daily operations and incident response (amoungst others), reduce time-to-resolution, and help our operations teams work more efficiently across Nutanix AHV clusters and Prism Central.

What we’re currently targeting:

• Faster Triage
  Use AI to analyze alert data, logs, and NCC output to prioritize incidents and highlight probable causes before a human even gets involved.
• Guided Resolution Paths
  AI suggests steps based on Nutanix KBs, internal SOPs, and historical incidents. Ideal for L1/L2 engineers who might not have deep Nutanix experience.
• Command and Script Recommendations
  Depending on whether it’s AHV or ESXi, the AI will suggest the right CLI commands or scripts based on the incident context. Helps eliminate trial-and-error.
• Consistent Documentation
  AI-generated summaries of the incident and resolution steps could be fed back into ServiceNow (or similar) for cleaner ticketing and audit trails.
• Reduced Escalation Overhead
  Long-term goal is to resolve more incidents at the first level with AI-backed guidance, so senior engineers can focus on actual engineering work.
• Automated Design Validation
  This is a big one we’re training the AI to validate environment designs and implementation plans against Nutanix best practices, supported hardware specs, and vendor reference architectures. The goal is to reduce the back-and-forth during peer reviews and catch issues before deployment.

It’s still early in the build, but the PoC architecture includes a GPU enabled Nutanix AHV Cluster, RHEL-based VMs running Ollama, LLaMA 3 models, FAISS for vector search, and Open WebUI for interaction. Each AI agent is scoped to a specific operational domain (Nutanix, NetApp, Commvault, Microsoft, etc.).

The Nutanix Design Assistant is also available in the public domain (openai framework), the PoC however is fully offline and uses offline Nutanix documentation, behavioural instruction sets and API integration for Prism Central…

If anyone’s experimenting with something similar, especially in regulated or air-gapped environments, I’d be interested in comparing notes.

Cheers,
John

Thanks for sharing ​@Jagoulden sounds like you have your hands full. Hope to learn more about it over the next few months.