A Kiro Power providing battle-tested guidance for deploying and training ML models on Amazon SageMaker AI. Covers inference endpoints, LLM fine-tuning, HyperPod clusters, Model Monitor, AutoML with AutoGluon, and SageMaker Python SDK v3 patterns — all derived from real deployment experience.
Integrates with the Amazon SageMaker AI MCP Server for HyperPod cluster management.
| Steering File | Topic |
|---|---|
model-customization.md |
Serverless model customization — SFT, DPO, RLVR, RLAIF trainers, reward functions, model evaluation (benchmarks, LLM-as-judge, custom scorers), deployment to SageMaker/Bedrock, supported models list |
inference-endpoints.md |
Real-time endpoint deployment — container selection, DJL LMI/vLLM, DLC env vars, multimodal models |
training-jobs.md |
Fine-tuning with custom scripts — QLoRA/LoRA, GPU vs Trainium, AWS recipes, Training Jobs |
hyperpod.md |
HyperPod cluster setup — EKS orchestration, Training Operator, Task Governance, resiliency |
hyperpod-inference.md |
HyperPod inference — JumpStart, custom models from S3/FSx, CLI/SDK/kubectl, autoscaling, KV caching |
model-monitor.md |
Model monitoring — Data Quality, Model Quality, Bias, Explainability, SDK v3 patterns and known bugs |
automl-autogluon.md |
AutoML — tabular, time series, multimodal, DLC images, SageMaker Pipelines |
sdk-v3-reference.md |
SDK v3 API reference — correct imports, image_uris, deployment patterns, invocation |
| Tool | Description |
|---|---|
manage_hyperpod_stacks |
Deploy, describe, and delete HyperPod clusters via CloudFormation |
manage_hyperpod_cluster_nodes |
List, describe, update, and delete cluster nodes |
-
Clone this repository:
git clone https://github.com/dgallitelli/kiro-power-for-sagemaker-ai.git
-
In Kiro, open the Powers panel (sidebar icon or Command Palette -> "Powers")
-
Click "Add Custom Power" -> "Local Directory"
-
Paste the absolute path to the cloned repo:
/path/to/kiro-power-for-sagemaker-ai -
Click "Add" — the power should appear in your Installed Powers list
-
In Kiro, open the Powers panel
-
Click "Add Custom Power" -> "Git Repository"
-
Paste:
https://github.com/dgallitelli/kiro-power-for-sagemaker-ai -
Click "Add"
Once installed, the power activates automatically when you work on SageMaker-related tasks. Kiro loads the relevant steering file based on your query:
- Ask about serverless model customization, fine-tuning with SFT/DPO/RLVR/RLAIF, or model evaluation -> loads
model-customization.md - Ask about deploying a model -> loads
inference-endpoints.md - Ask about custom training scripts or Training Jobs -> loads
training-jobs.md - Ask about HyperPod -> loads
hyperpod.mdorhyperpod-inference.md - Ask about model monitoring -> loads
model-monitor.md - Ask about AutoGluon -> loads
automl-autogluon.md
You can also explicitly reference steering files in chat using # context.
Apache-2.0