Nutanix Artificial Intelligence Certification Exam Syllabus

• Identify the installation prerequisites
• Identify the installation limitations
• Cite the installation procedure
• Describe the core fundamental components of NAI architecture

- Install Nutanix Enterprise AI components

• Compare and contrast the installation process for NKP (including app catalog) and non NKP environments
• Recognize version compatibility between pre-requisite and NAI components
• Perform a dark site installation
• Configure storage classes

• Identify or implement an FQDN for the NAI installation
• Ensure that the FQDN has a secure certificate
• Validate successful login to UI

• Differentiate between the user and administrator roles
• Identify the user management operations that can be performed as an administrator
• Given specific scenarios, leverage roles to limit privileges for target users

- Import Large Language Models (LLMs)

• Recognize the methods and repos available for importing
• Obtain repo keys for HuggingFace and/or NVIDIA NGC
• Recognize where to add repo keys in the UI for usage
• Explain the manual import process and the requirements

- Create endpoints

• Determine a downloaded model to expose via an endpoint
• Determine the number and type of GPUs required for a selected model
• Determine the number of instances required to achieve a certain throughput
• Determine vCPU/memory and inference engine for optimization scenarios

- Create and apply keys for each API endpoint

• Identify the locations to generate and manage API keys
• Identify where to view API keys in an endpoint
• Deactivate an API key
• Add an API key to an existing endpoint

- Deliver endpoints to the consumer

• Identify the endpoint URI and model-specific parameters and the API key to be shared with consumers
• Identify tool calling vs non tool calling API curl commands

• Determine where to get the sample request in the NAI application
• Explain the elements in the sample request and the elements required for the OpenAI compliant application configuration
• Recognize the different endpoint types and choose the correct one for a given application

• Determine the observability metrics for performance evaluation
• Determine possible changes in resource allocation to remedy latency and throughput issues

• Determine where and how to view the top 5 API Keys being used
• Locate the endpoint dashboard and view assigned API keys
• Recognize when to deactivate API keys
• Review and interpret audit events

• Determine the prompt input and the LLM output per endpoint to evaluate accuracy through human feedback
• Determine techniques and models that can be used to improve the output quality
• Apply guardrails to improve safety
• Apply rerank models to achieve desired results

• Determine where to view infrastructure performance
• Recognize how to filter by GPU nodes and review resulting GPU utilization graph to determine which GPUs are being heavily used
• Determine if an endpoint is using GPU
• Recognize which type of GPU an endpoint is using.
• Determine if endpoint is using CPU-based acceleration or not

- Remediate health check failures on the cluster

• Debug a cluster health fail visible on NAI UI
• Recognize the different components that can cause health check failures
• Analyze Kubernetes NAI system resources to address health check failures
• Determine which layer of the stack is causing the health check failure
• Based on a health check failure diagnosis, determine an appropriate course of action

- Troubleshoot model import and endpoint creation

• Identify the failure scenarios where model download fails due to misconfigurations and/or connectivity (e.g., prevalidated, custom, and/or restricted networks)
  - Troubleshoot CSI driver connectivity
  - Ensure model EULA was accepted on HuggingFace (Llama models)
  - Ensure HuggingFace or NVIDIA token is valid
• Determine available allocatable compute resources (e.g., CPU, memory, GPUs, taints) that could prevent endpoints from being scheduled
• Recognize if all prerequisites were successfully installed (e.g. Kserve)
• Diagnose the cause of container images failing to be downloaded or be stored on Kubernetes nodes

• Differentiate between model and endpoint types to be consumed by the application
• Recognize the purpose and use case for integrating the various types of models
• Issue a simple query to the OpenAI-compatible NAI API endpoints using Python or Curl
• Investigate and address application integration issues

• Identify the latency and number of API requests per endpoint associated with application
• Describe how to correlate the application with NAI endpoint metrics