rh-ai-engineer

Analyze AI model performance, GPU utilization, and cluster health on OpenShift AI. Use when: - "How is my model performing?" - "What GPUs are available in the cluster?" - "Show me inference latency for Llama" - "Check OpenShift cluster health metrics" - "Trace a slow inference request" - "Correlate errors across my inference stack" Query-driven, read-only analysis. Routes to the appropriate observability domain based on user intent. NOT for deploying models (use /model-deploy). NOT for debugging failed deployments (use /debug-inference).

Troubleshoot failed or slow InferenceService deployments on OpenShift AI. Use when: - "My InferenceService won't start" - "Model deployment is stuck" - "Inference endpoint returns errors" - "Model is slow / high latency" - "GPU scheduling failed for my model" Progressive diagnosis: status conditions, events, pod logs, GPU health, and observability analysis. NOT for deploying models (use /model-deploy). NOT for creating runtimes (use /serving-runtime-config).

Create and configure Data Science Projects on OpenShift AI with namespace setup, S3 data connections, pipeline server, and model serving enablement. Use when: - "Create a data science project" - "Set up a new namespace for ML work" - "Add an S3 data connection to my project" - "Configure the pipeline server" - "Enable model serving on my project" Bootstraps an RHOAI Data Science Project with proper labels, data connections, pipeline infrastructure, and model serving configuration. NOT for deploying models (use /model-deploy). NOT for creating workbenches (use /workbench-manage). NOT for managing pipelines after setup (use /pipeline-manage).

Configure TrustyAI Guardrails Orchestrator for LLM input/output content safety on OpenShift AI. Use when: - "Add guardrails to my LLM endpoint" - "Set up content safety for my model" - "Configure PII detection on my inference endpoint" - "Block prompt injection attacks" - "I need a guarded endpoint for my deployed model" Handles GuardrailsOrchestrator CR deployment, detector configuration (content safety, PII, prompt injection, toxicity), orchestration policies, and guarded endpoint validation. NOT for deploying models (use /model-deploy first). NOT for bias/drift monitoring (use /model-monitor). NOT for infrastructure observability (use /ai-observability).

Deploy AI/ML models on OpenShift AI using KServe with vLLM, NVIDIA NIM, or Caikit+TGIS runtimes. Use when: - "Deploy Llama 3 on my cluster" - "Set up a vLLM inference endpoint" - "Deploy a model with NIM" - "Create an InferenceService for Granite" - "I need to serve a model on OpenShift AI" Handles runtime selection, GPU validation, InferenceService CR creation, and rollout monitoring. NOT for NIM platform setup (use /nim-setup first). NOT for custom runtime creation (use /serving-runtime-config).

Configure TrustyAI model monitoring for bias detection and data drift on deployed InferenceServices. Use when: - "Monitor my model for bias" - "Set up drift detection on my inference endpoint" - "Configure TrustyAI for my deployed model" - "Check if my model has fairness issues" - "I need SPD / DIR metrics for my model" Handles TrustyAIService deployment, bias metric configuration (SPD, DIR), drift metric configuration (MeanShift, FourierMMD, KS-Test, Jensen-Shannon), threshold tuning, and monitoring validation. NOT for deploying models (use /model-deploy first). NOT for input/output content safety guardrails (use /guardrails-config). NOT for infrastructure-level observability (use /ai-observability).

Register, version, and manage ML models in the OpenShift AI Model Registry. Browse the Model Catalog, track model metadata, and promote models across environments. Use when: - "Register a new model in the registry" - "List registered models" - "What versions exist for my model?" - "Promote a model from dev to production" - "Show model artifacts and storage URIs" Handles model registration, versioning, metadata management, artifact tracking, and cross-environment promotion. NOT for deploying models (use /model-deploy). NOT for model performance monitoring (use /ai-observability).

Configure NVIDIA NIM platform on OpenShift AI for optimized model inference. Use when: - "Set up NIM on my cluster" - "Configure NGC credentials for NIM" - "I want to deploy a NIM model but haven't set up the platform" - "Create the NIM Account CR" One-time prerequisite before deploying models with NVIDIA NIM runtime via /model-deploy. NOT for deploying models (use /model-deploy instead). NOT for vLLM or Caikit deployments (NIM-specific only).

Create, run, schedule, and monitor Data Science Pipelines (Kubeflow Pipelines 2.0) on OpenShift AI. Use when: - "Run a pipeline in my project" - "Schedule a recurring pipeline" - "Check my pipeline run status" - "List pipeline runs and their logs" - "Set up the pipeline server" - "Delete a pipeline or pipeline run" Handles pipeline server setup, pipeline run submission from YAML, scheduling recurring runs, monitoring execution, and viewing step logs. NOT for creating data science projects (use /ds-project-setup). NOT for deploying models (use /model-deploy). NOT for model training jobs (use training skills).

Configure custom ServingRuntime CRs on OpenShift AI for model serving frameworks not covered by built-in runtimes. Use when: - "Create a custom serving runtime" - "I need a runtime for ONNX / Triton / custom framework" - "Customize vLLM runtime parameters" - "What serving runtimes are available?" - "Add a custom container image for model serving" Handles listing existing runtimes, creating new ServingRuntime CRs, and validating compatibility with target models. NOT for deploying models (use /model-deploy after runtime is configured). NOT for NIM platform setup (use /nim-setup).

Create and manage Jupyter notebook workbenches on OpenShift AI with image selection, resource configuration, PVC storage, and lifecycle management. Use when: - "Create a notebook workbench" - "Spin up a Jupyter environment for data science" - "Start / stop my workbench" - "What notebook images are available?" - "Delete a workbench I no longer need" Handles Notebook CR lifecycle: create with configurable images and resources, start/stop, attach storage, and delete with data loss warnings. NOT for deploying models (use /model-deploy). NOT for creating projects (use /ds-project-setup). NOT for managing pipelines (use /pipeline-manage).

OpenShift MCP server for Kubernetes resource CRUD, pod management, logs, and events

AI Observability MCP server for vLLM model performance analysis, GPU monitoring, OpenShift metrics, and distributed tracing. Deployed on-cluster from quay.io/ecosystem-appeng/aiobs-mcp-server. See: https://github.com/rh-ai-quickstart/ai-observability-summarizer

Red Hat OpenShift AI MCP server for high-level RHOAI operations: Data Science Projects, model serving (InferenceService deploy/list/get), serving runtimes, data connections, workbenches, pipelines, storage, training, and model registry. See: https://github.com/opendatahub-io/rhoai-mcp