Ml Ops

MLOps

Activate this skill when deploying models or managing ML infrastructure.

When to Use

• Deploying models to production endpoints
• Setting up experiment tracking (MLflow, W&B)
• Building model registries and versioning
• Implementing A/B testing for models
• Monitoring model drift and performance

Core Tools

• MLflow: Experiment tracking, model registry
• Weights & Biases: Experiment visualization
• BentoML/Ray Serve: Model serving
• DVC: Data and model versioning
• Evidently AI: Model monitoring

Patterns

import mlflow

mlflow.set_experiment("classification_v2")
with mlflow.start_run():
    mlflow.log_params({"lr": 0.001, "epochs": 50})
    mlflow.log_metrics({"accuracy": 0.94, "f1": 0.91})
    mlflow.sklearn.log_model(model, "model")

Rules

• Version everything: code, data, models, configs
• Automate training pipelines (no manual steps)
• Monitor prediction distributions for drift
• Implement rollback mechanisms for model updates
• Log all experiments, even failed ones

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: ihatesea69
• Source: ihatesea69/kiro-kit
• License: MIT
• Homepage: https://www.npmjs.com/package/kiro-kit

Install and usage instructions live in the source repository linked above.